#pragma once

#include "logging.h"
#include "device_functions-generated.h"
#include "model_interface.h"
#include "raii_wrapper.h"
#include "model.h"
#include "macros.h"
#include "jagged.h"
#include <algorithm>
#include <deque>
#include <fstream>
#include <iostream>
#include <string>
#include <unordered_map>
#include <math.h>
#include <iomanip>


void conv2d_bias_few_channels_0(
  void*,
  void*,
  void*,

  void*,

  uint8_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int,
  int,
  int,
  int,
  int,
  int,
  cudaStream_t
);

void invoke_fused_elementwise_425(void* output0, const void* input0,const void* input1,   int64_t n_elements, cudaStream_t stream);
    

    cudaError_t layernorm_3(void* output,
                   void* input,
                   const void* gamma,
                   const void* beta,
                   int m,
                   int n,
                   const float eps,
                   cudaStream_t stream);
    

void gemm_rcr_bias_permute_5(
  void*,
  void*,
  void*,
  void*,
  uint8_t*,

    int,


  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,

  cudaStream_t
);

void split_7(
    void *[] /*outputs*/,
    int64_t **[] /*output_shapes*/,
    const bool [] /*output_masks*/,
    const void * /*input*/,
    const int64_t * /*input_shape*/,
    int64_t /*real_num_splits*/,
    int64_t /*all_num_splits*/,
    int64_t [] /*split_sizes*/,
    int64_t /*split_dim*/,
    int64_t /*rank*/,
    cudaStream_t stream
);

void ait_reshape_8(
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*
);

    void mem_eff_attention_11(void* output,
                   void* query,
                   void* key,
                   void* value,
                   int64_t* batch_size,
                   int64_t* seq_len_kv,
                   int64_t* seq_len_q,
                   int num_heads,
                   int head_size,
                   int head_size_v,
                   float p_dropout,
                   float softmax_scale,
                   bool is_causal,
                   bool fixed_seq_length_kv,
                   int32_t* lengths_kv,
                   bool fixed_seq_length_q,
                   int32_t* lengths_q,
                   void* workspace,
                   cudaStream_t stream);
    

void gemm_rcr_bias_add_13(
  void*,
  void*,
  void*,
  void*,

  void*,
  uint8_t*,

    int,


  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,

  cudaStream_t
);

    cudaError_t layernorm_15(void* output,
                   void* input,
                   const void* gamma,
                   const void* beta,
                   int m,
                   int n,
                   const float eps,
                   cudaStream_t stream);
    

void gemm_rcr_bias_gelu_16(
  void*,
  void*,
  void*,
  void*,
  uint8_t*,

    int,


  int64_t*,

  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,

  int64_t*,

  cudaStream_t
);

void gemm_rcr_bias_add_17(
  void*,
  void*,
  void*,
  void*,

  void*,
  uint8_t*,

    int,


  int64_t*,

  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,

  int64_t*,

  cudaStream_t
);

void gemm_rcr_bias_409(
  void*,
  void*,
  void*,
  void*,
  uint8_t*,

    int,


  int64_t*,

  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,

  int64_t*,

  cudaStream_t
);

void permute_411(
  void*,
  const void*,

  int64_t*,

  int64_t*,

  int64_t*,

  int64_t*,

  int64_t*,

  const int*,
  cudaStream_t
);

void split_412(
    void *[] /*outputs*/,
    int64_t **[] /*output_shapes*/,
    const bool [] /*output_masks*/,
    const void * /*input*/,
    const int64_t * /*input_shape*/,
    int64_t /*real_num_splits*/,
    int64_t /*all_num_splits*/,
    int64_t [] /*split_sizes*/,
    int64_t /*split_dim*/,
    int64_t /*rank*/,
    cudaStream_t stream
);

void ait_squeeze_413(
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*
);

    void mem_eff_attention_418(void* output,
                   void* query,
                   void* key,
                   void* value,
                   int64_t* batch_size,
                   int64_t* seq_len_kv,
                   int64_t* seq_len_q,
                   int num_heads,
                   int head_size,
                   int head_size_v,
                   float p_dropout,
                   float softmax_scale,
                   bool is_causal,
                   bool fixed_seq_length_kv,
                   int32_t* lengths_kv,
                   bool fixed_seq_length_q,
                   int32_t* lengths_q,
                   void* workspace,
                   cudaStream_t stream);
    

void ait_reshape_426(
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*,
  int64_t*
);

void gemm_rcr_bias_421(
  void*,
  void*,
  void*,
  void*,
  uint8_t*,

    int,


  int64_t*,

  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,

  int64_t*,

  cudaStream_t
);

    cudaError_t layernorm_422(void* output,
                   void* input,
                   const void* gamma,
                   const void* beta,
                   int m,
                   int n,
                   const float eps,
                   cudaStream_t stream);
    

void gemm_rcr_bias_gelu_423(
  void*,
  void*,
  void*,
  void*,
  uint8_t*,

    int,


  int64_t*,

  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,

  int64_t*,

  cudaStream_t
);

void gemm_rcr_bias_add_424(
  void*,
  void*,
  void*,
  void*,

  void*,
  uint8_t*,

    int,


  int64_t*,

  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,


  int64_t*,

  int64_t*,

  int64_t*,

  cudaStream_t
);

namespace ait {

// Model is the class that actually performs inference. It owns memory for
// intermediate tensors and dynamic dimensions. Constants are owned by
// the model's owning container object, and input/output memory is owned
// by the user.
// Once an inference run has started, it is not safe to re-use the Model
// until the run has finished!
class Model : public ModelBase<Model> {
  
  

  public:
    Model(
        size_t blob_size,
        size_t workspace_size,
        size_t unique_workspace_size,
        size_t num_inputs,
        size_t num_outputs,
        size_t num_unbound_constants,
        uint8_t* constants,
        AITemplateAllocator& allocator)
        : ModelBase(
            blob_size,
            workspace_size,
            unique_workspace_size,
            num_inputs,
            num_outputs,
            num_unbound_constants,
            constants,
            allocator) {
         constant_name_to_ptr_["patch_embed_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&patch_embed_proj_weight));
     constant_name_to_ptr_["patch_embed_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&patch_embed_proj_bias));
     constant_name_to_ptr_["pos_emb_pos_emb"] = const_cast<const void**>(reinterpret_cast<void**>(&pos_emb_pos_emb));
     constant_name_to_ptr_["encoder_layers_0_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_ln1_weight));
     constant_name_to_ptr_["encoder_layers_0_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_ln1_bias));
     constant_name_to_ptr_["encoder_layers_0_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_0_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_0_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_0_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_0_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_ln2_weight));
     constant_name_to_ptr_["encoder_layers_0_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_ln2_bias));
     constant_name_to_ptr_["encoder_layers_0_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_0_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_0_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_0_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_0_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_1_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_ln1_weight));
     constant_name_to_ptr_["encoder_layers_1_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_ln1_bias));
     constant_name_to_ptr_["encoder_layers_1_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_1_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_1_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_1_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_1_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_ln2_weight));
     constant_name_to_ptr_["encoder_layers_1_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_ln2_bias));
     constant_name_to_ptr_["encoder_layers_1_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_1_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_1_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_1_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_1_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_2_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_ln1_weight));
     constant_name_to_ptr_["encoder_layers_2_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_ln1_bias));
     constant_name_to_ptr_["encoder_layers_2_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_2_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_2_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_2_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_2_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_ln2_weight));
     constant_name_to_ptr_["encoder_layers_2_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_ln2_bias));
     constant_name_to_ptr_["encoder_layers_2_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_2_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_2_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_2_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_2_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_3_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_ln1_weight));
     constant_name_to_ptr_["encoder_layers_3_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_ln1_bias));
     constant_name_to_ptr_["encoder_layers_3_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_3_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_3_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_3_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_3_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_ln2_weight));
     constant_name_to_ptr_["encoder_layers_3_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_ln2_bias));
     constant_name_to_ptr_["encoder_layers_3_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_3_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_3_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_3_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_3_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_4_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_ln1_weight));
     constant_name_to_ptr_["encoder_layers_4_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_ln1_bias));
     constant_name_to_ptr_["encoder_layers_4_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_4_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_4_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_4_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_4_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_ln2_weight));
     constant_name_to_ptr_["encoder_layers_4_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_ln2_bias));
     constant_name_to_ptr_["encoder_layers_4_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_4_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_4_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_4_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_4_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_5_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_ln1_weight));
     constant_name_to_ptr_["encoder_layers_5_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_ln1_bias));
     constant_name_to_ptr_["encoder_layers_5_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_5_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_5_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_5_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_5_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_ln2_weight));
     constant_name_to_ptr_["encoder_layers_5_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_ln2_bias));
     constant_name_to_ptr_["encoder_layers_5_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_5_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_5_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_5_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_5_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_6_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_ln1_weight));
     constant_name_to_ptr_["encoder_layers_6_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_ln1_bias));
     constant_name_to_ptr_["encoder_layers_6_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_6_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_6_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_6_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_6_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_ln2_weight));
     constant_name_to_ptr_["encoder_layers_6_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_ln2_bias));
     constant_name_to_ptr_["encoder_layers_6_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_6_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_6_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_6_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_6_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_7_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_ln1_weight));
     constant_name_to_ptr_["encoder_layers_7_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_ln1_bias));
     constant_name_to_ptr_["encoder_layers_7_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_7_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_7_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_7_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_7_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_ln2_weight));
     constant_name_to_ptr_["encoder_layers_7_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_ln2_bias));
     constant_name_to_ptr_["encoder_layers_7_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_7_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_7_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_7_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_7_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_8_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_ln1_weight));
     constant_name_to_ptr_["encoder_layers_8_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_ln1_bias));
     constant_name_to_ptr_["encoder_layers_8_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_8_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_8_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_8_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_8_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_ln2_weight));
     constant_name_to_ptr_["encoder_layers_8_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_ln2_bias));
     constant_name_to_ptr_["encoder_layers_8_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_8_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_8_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_8_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_8_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_9_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_ln1_weight));
     constant_name_to_ptr_["encoder_layers_9_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_ln1_bias));
     constant_name_to_ptr_["encoder_layers_9_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_9_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_9_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_9_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_9_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_ln2_weight));
     constant_name_to_ptr_["encoder_layers_9_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_ln2_bias));
     constant_name_to_ptr_["encoder_layers_9_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_9_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_9_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_9_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_9_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_10_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_ln1_weight));
     constant_name_to_ptr_["encoder_layers_10_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_ln1_bias));
     constant_name_to_ptr_["encoder_layers_10_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_10_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_10_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_10_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_10_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_ln2_weight));
     constant_name_to_ptr_["encoder_layers_10_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_ln2_bias));
     constant_name_to_ptr_["encoder_layers_10_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_10_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_10_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_10_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_10_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_11_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_ln1_weight));
     constant_name_to_ptr_["encoder_layers_11_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_ln1_bias));
     constant_name_to_ptr_["encoder_layers_11_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_11_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_11_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_11_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_11_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_ln2_weight));
     constant_name_to_ptr_["encoder_layers_11_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_ln2_bias));
     constant_name_to_ptr_["encoder_layers_11_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_11_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_11_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_11_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_11_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_12_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_ln1_weight));
     constant_name_to_ptr_["encoder_layers_12_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_ln1_bias));
     constant_name_to_ptr_["encoder_layers_12_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_12_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_12_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_12_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_12_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_ln2_weight));
     constant_name_to_ptr_["encoder_layers_12_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_ln2_bias));
     constant_name_to_ptr_["encoder_layers_12_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_12_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_12_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_12_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_12_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_13_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_ln1_weight));
     constant_name_to_ptr_["encoder_layers_13_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_ln1_bias));
     constant_name_to_ptr_["encoder_layers_13_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_13_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_13_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_13_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_13_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_ln2_weight));
     constant_name_to_ptr_["encoder_layers_13_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_ln2_bias));
     constant_name_to_ptr_["encoder_layers_13_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_13_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_13_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_13_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_13_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_14_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_ln1_weight));
     constant_name_to_ptr_["encoder_layers_14_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_ln1_bias));
     constant_name_to_ptr_["encoder_layers_14_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_14_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_14_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_14_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_14_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_ln2_weight));
     constant_name_to_ptr_["encoder_layers_14_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_ln2_bias));
     constant_name_to_ptr_["encoder_layers_14_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_14_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_14_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_14_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_14_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_15_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_ln1_weight));
     constant_name_to_ptr_["encoder_layers_15_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_ln1_bias));
     constant_name_to_ptr_["encoder_layers_15_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_15_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_15_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_15_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_15_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_ln2_weight));
     constant_name_to_ptr_["encoder_layers_15_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_ln2_bias));
     constant_name_to_ptr_["encoder_layers_15_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_15_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_15_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_15_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_15_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_16_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_ln1_weight));
     constant_name_to_ptr_["encoder_layers_16_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_ln1_bias));
     constant_name_to_ptr_["encoder_layers_16_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_16_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_16_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_16_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_16_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_ln2_weight));
     constant_name_to_ptr_["encoder_layers_16_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_ln2_bias));
     constant_name_to_ptr_["encoder_layers_16_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_16_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_16_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_16_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_16_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_17_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_ln1_weight));
     constant_name_to_ptr_["encoder_layers_17_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_ln1_bias));
     constant_name_to_ptr_["encoder_layers_17_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_17_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_17_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_17_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_17_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_ln2_weight));
     constant_name_to_ptr_["encoder_layers_17_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_ln2_bias));
     constant_name_to_ptr_["encoder_layers_17_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_17_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_17_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_17_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_17_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_18_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_ln1_weight));
     constant_name_to_ptr_["encoder_layers_18_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_ln1_bias));
     constant_name_to_ptr_["encoder_layers_18_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_18_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_18_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_18_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_18_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_ln2_weight));
     constant_name_to_ptr_["encoder_layers_18_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_ln2_bias));
     constant_name_to_ptr_["encoder_layers_18_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_18_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_18_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_18_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_18_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_19_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_ln1_weight));
     constant_name_to_ptr_["encoder_layers_19_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_ln1_bias));
     constant_name_to_ptr_["encoder_layers_19_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_19_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_19_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_19_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_19_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_ln2_weight));
     constant_name_to_ptr_["encoder_layers_19_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_ln2_bias));
     constant_name_to_ptr_["encoder_layers_19_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_19_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_19_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_19_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_19_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_20_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_ln1_weight));
     constant_name_to_ptr_["encoder_layers_20_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_ln1_bias));
     constant_name_to_ptr_["encoder_layers_20_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_20_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_20_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_20_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_20_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_ln2_weight));
     constant_name_to_ptr_["encoder_layers_20_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_ln2_bias));
     constant_name_to_ptr_["encoder_layers_20_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_20_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_20_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_20_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_20_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_21_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_ln1_weight));
     constant_name_to_ptr_["encoder_layers_21_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_ln1_bias));
     constant_name_to_ptr_["encoder_layers_21_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_21_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_21_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_21_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_21_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_ln2_weight));
     constant_name_to_ptr_["encoder_layers_21_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_ln2_bias));
     constant_name_to_ptr_["encoder_layers_21_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_21_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_21_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_21_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_21_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_22_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_ln1_weight));
     constant_name_to_ptr_["encoder_layers_22_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_ln1_bias));
     constant_name_to_ptr_["encoder_layers_22_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_22_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_22_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_22_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_22_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_ln2_weight));
     constant_name_to_ptr_["encoder_layers_22_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_ln2_bias));
     constant_name_to_ptr_["encoder_layers_22_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_22_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_22_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_22_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_22_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_23_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_ln1_weight));
     constant_name_to_ptr_["encoder_layers_23_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_ln1_bias));
     constant_name_to_ptr_["encoder_layers_23_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_23_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_23_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_23_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_23_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_ln2_weight));
     constant_name_to_ptr_["encoder_layers_23_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_ln2_bias));
     constant_name_to_ptr_["encoder_layers_23_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_23_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_23_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_23_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_23_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_24_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_ln1_weight));
     constant_name_to_ptr_["encoder_layers_24_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_ln1_bias));
     constant_name_to_ptr_["encoder_layers_24_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_24_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_24_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_24_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_24_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_ln2_weight));
     constant_name_to_ptr_["encoder_layers_24_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_ln2_bias));
     constant_name_to_ptr_["encoder_layers_24_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_24_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_24_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_24_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_24_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_25_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_ln1_weight));
     constant_name_to_ptr_["encoder_layers_25_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_ln1_bias));
     constant_name_to_ptr_["encoder_layers_25_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_25_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_25_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_25_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_25_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_ln2_weight));
     constant_name_to_ptr_["encoder_layers_25_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_ln2_bias));
     constant_name_to_ptr_["encoder_layers_25_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_25_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_25_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_25_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_25_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_layers_26_ln1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_ln1_weight));
     constant_name_to_ptr_["encoder_layers_26_ln1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_ln1_bias));
     constant_name_to_ptr_["encoder_layers_26_mha_qkv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_mha_qkv_weight));
     constant_name_to_ptr_["encoder_layers_26_mha_qkv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_mha_qkv_bias));
     constant_name_to_ptr_["encoder_layers_26_mha_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_mha_proj_weight));
     constant_name_to_ptr_["encoder_layers_26_mha_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_mha_proj_bias));
     constant_name_to_ptr_["encoder_layers_26_ln2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_ln2_weight));
     constant_name_to_ptr_["encoder_layers_26_ln2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_ln2_bias));
     constant_name_to_ptr_["encoder_layers_26_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_mlp_fc1_weight));
     constant_name_to_ptr_["encoder_layers_26_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_mlp_fc1_bias));
     constant_name_to_ptr_["encoder_layers_26_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_mlp_fc2_weight));
     constant_name_to_ptr_["encoder_layers_26_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_layers_26_mlp_fc2_bias));
     constant_name_to_ptr_["encoder_ln_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_ln_weight));
     constant_name_to_ptr_["encoder_ln_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&encoder_ln_bias));
     constant_name_to_ptr_["pool_kv_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&pool_kv_weight));
     constant_name_to_ptr_["pool_kv_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&pool_kv_bias));
     constant_name_to_ptr_["pool_proj_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&pool_proj_weight));
     constant_name_to_ptr_["pool_proj_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&pool_proj_bias));
     constant_name_to_ptr_["pool_ln_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&pool_ln_weight));
     constant_name_to_ptr_["pool_ln_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&pool_ln_bias));
     constant_name_to_ptr_["pool_mlp_fc1_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&pool_mlp_fc1_weight));
     constant_name_to_ptr_["pool_mlp_fc1_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&pool_mlp_fc1_bias));
     constant_name_to_ptr_["pool_mlp_fc2_weight"] = const_cast<const void**>(reinterpret_cast<void**>(&pool_mlp_fc2_weight));
     constant_name_to_ptr_["pool_mlp_fc2_bias"] = const_cast<const void**>(reinterpret_cast<void**>(&pool_mlp_fc2_bias));
    auto* blob_ptr = static_cast<uint8_t*>(blob_.get());
        conv2d_bias_few_channels_0_0 = reinterpret_cast<decltype(conv2d_bias_few_channels_0_0)>(blob_ptr + 0);
    elementwise_2_0 = reinterpret_cast<decltype(elementwise_2_0)>(blob_ptr + 40310784);
    reshape_4_0 = reinterpret_cast<decltype(reshape_4_0)>(blob_ptr + 53747712);
    reshape_6_0 = reinterpret_cast<decltype(reshape_6_0)>(blob_ptr + 0);
    split_7_0 = reinterpret_cast<decltype(split_7_0)>(blob_ptr + 53747712);
    split_7_1 = reinterpret_cast<decltype(split_7_1)>(blob_ptr + 67184640);
    split_7_2 = reinterpret_cast<decltype(split_7_2)>(blob_ptr + 80621568);
    mem_eff_attention_11_0 = reinterpret_cast<decltype(mem_eff_attention_11_0)>(blob_ptr + 0);
    reshape_14_0 = reinterpret_cast<decltype(reshape_14_0)>(blob_ptr + 53747712);
    layernorm_15_0 = reinterpret_cast<decltype(layernorm_15_0)>(blob_ptr + 67184640);
    gemm_rcr_bias_gelu_16_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_16_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_17_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_17_0)>(blob_ptr + 67184640);
    reshape_19_0 = reinterpret_cast<decltype(reshape_19_0)>(blob_ptr + 40310784);
    reshape_21_0 = reinterpret_cast<decltype(reshape_21_0)>(blob_ptr + 0);
    split_22_0 = reinterpret_cast<decltype(split_22_0)>(blob_ptr + 40310784);
    split_22_1 = reinterpret_cast<decltype(split_22_1)>(blob_ptr + 80621568);
    split_22_2 = reinterpret_cast<decltype(split_22_2)>(blob_ptr + 53747712);
    mem_eff_attention_26_0 = reinterpret_cast<decltype(mem_eff_attention_26_0)>(blob_ptr + 0);
    reshape_29_0 = reinterpret_cast<decltype(reshape_29_0)>(blob_ptr + 50201856);
    layernorm_30_0 = reinterpret_cast<decltype(layernorm_30_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_31_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_31_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_32_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_32_0)>(blob_ptr + 63638784);
    reshape_34_0 = reinterpret_cast<decltype(reshape_34_0)>(blob_ptr + 40310784);
    reshape_36_0 = reinterpret_cast<decltype(reshape_36_0)>(blob_ptr + 0);
    split_37_0 = reinterpret_cast<decltype(split_37_0)>(blob_ptr + 77075712);
    split_37_1 = reinterpret_cast<decltype(split_37_1)>(blob_ptr + 40310784);
    split_37_2 = reinterpret_cast<decltype(split_37_2)>(blob_ptr + 90512640);
    mem_eff_attention_41_0 = reinterpret_cast<decltype(mem_eff_attention_41_0)>(blob_ptr + 0);
    reshape_44_0 = reinterpret_cast<decltype(reshape_44_0)>(blob_ptr + 50201856);
    layernorm_45_0 = reinterpret_cast<decltype(layernorm_45_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_46_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_46_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_47_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_47_0)>(blob_ptr + 63638784);
    reshape_49_0 = reinterpret_cast<decltype(reshape_49_0)>(blob_ptr + 40310784);
    reshape_51_0 = reinterpret_cast<decltype(reshape_51_0)>(blob_ptr + 0);
    split_52_0 = reinterpret_cast<decltype(split_52_0)>(blob_ptr + 90512640);
    split_52_1 = reinterpret_cast<decltype(split_52_1)>(blob_ptr + 77075712);
    split_52_2 = reinterpret_cast<decltype(split_52_2)>(blob_ptr + 40310784);
    mem_eff_attention_56_0 = reinterpret_cast<decltype(mem_eff_attention_56_0)>(blob_ptr + 0);
    reshape_59_0 = reinterpret_cast<decltype(reshape_59_0)>(blob_ptr + 50201856);
    layernorm_60_0 = reinterpret_cast<decltype(layernorm_60_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_61_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_61_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_62_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_62_0)>(blob_ptr + 63638784);
    reshape_64_0 = reinterpret_cast<decltype(reshape_64_0)>(blob_ptr + 40310784);
    reshape_66_0 = reinterpret_cast<decltype(reshape_66_0)>(blob_ptr + 0);
    split_67_0 = reinterpret_cast<decltype(split_67_0)>(blob_ptr + 40310784);
    split_67_1 = reinterpret_cast<decltype(split_67_1)>(blob_ptr + 77075712);
    split_67_2 = reinterpret_cast<decltype(split_67_2)>(blob_ptr + 90512640);
    mem_eff_attention_71_0 = reinterpret_cast<decltype(mem_eff_attention_71_0)>(blob_ptr + 0);
    reshape_74_0 = reinterpret_cast<decltype(reshape_74_0)>(blob_ptr + 50201856);
    layernorm_75_0 = reinterpret_cast<decltype(layernorm_75_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_76_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_76_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_77_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_77_0)>(blob_ptr + 63638784);
    reshape_79_0 = reinterpret_cast<decltype(reshape_79_0)>(blob_ptr + 40310784);
    reshape_81_0 = reinterpret_cast<decltype(reshape_81_0)>(blob_ptr + 0);
    split_82_0 = reinterpret_cast<decltype(split_82_0)>(blob_ptr + 40310784);
    split_82_1 = reinterpret_cast<decltype(split_82_1)>(blob_ptr + 77075712);
    split_82_2 = reinterpret_cast<decltype(split_82_2)>(blob_ptr + 90512640);
    mem_eff_attention_86_0 = reinterpret_cast<decltype(mem_eff_attention_86_0)>(blob_ptr + 0);
    reshape_89_0 = reinterpret_cast<decltype(reshape_89_0)>(blob_ptr + 50201856);
    layernorm_90_0 = reinterpret_cast<decltype(layernorm_90_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_91_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_91_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_92_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_92_0)>(blob_ptr + 63638784);
    reshape_94_0 = reinterpret_cast<decltype(reshape_94_0)>(blob_ptr + 40310784);
    reshape_96_0 = reinterpret_cast<decltype(reshape_96_0)>(blob_ptr + 0);
    split_97_0 = reinterpret_cast<decltype(split_97_0)>(blob_ptr + 77075712);
    split_97_1 = reinterpret_cast<decltype(split_97_1)>(blob_ptr + 90512640);
    split_97_2 = reinterpret_cast<decltype(split_97_2)>(blob_ptr + 40310784);
    mem_eff_attention_101_0 = reinterpret_cast<decltype(mem_eff_attention_101_0)>(blob_ptr + 0);
    reshape_104_0 = reinterpret_cast<decltype(reshape_104_0)>(blob_ptr + 50201856);
    layernorm_105_0 = reinterpret_cast<decltype(layernorm_105_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_106_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_106_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_107_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_107_0)>(blob_ptr + 63638784);
    reshape_109_0 = reinterpret_cast<decltype(reshape_109_0)>(blob_ptr + 40310784);
    reshape_111_0 = reinterpret_cast<decltype(reshape_111_0)>(blob_ptr + 0);
    split_112_0 = reinterpret_cast<decltype(split_112_0)>(blob_ptr + 40310784);
    split_112_1 = reinterpret_cast<decltype(split_112_1)>(blob_ptr + 77075712);
    split_112_2 = reinterpret_cast<decltype(split_112_2)>(blob_ptr + 90512640);
    mem_eff_attention_116_0 = reinterpret_cast<decltype(mem_eff_attention_116_0)>(blob_ptr + 0);
    reshape_119_0 = reinterpret_cast<decltype(reshape_119_0)>(blob_ptr + 50201856);
    layernorm_120_0 = reinterpret_cast<decltype(layernorm_120_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_121_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_121_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_122_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_122_0)>(blob_ptr + 63638784);
    reshape_124_0 = reinterpret_cast<decltype(reshape_124_0)>(blob_ptr + 40310784);
    reshape_126_0 = reinterpret_cast<decltype(reshape_126_0)>(blob_ptr + 0);
    split_127_0 = reinterpret_cast<decltype(split_127_0)>(blob_ptr + 77075712);
    split_127_1 = reinterpret_cast<decltype(split_127_1)>(blob_ptr + 40310784);
    split_127_2 = reinterpret_cast<decltype(split_127_2)>(blob_ptr + 90512640);
    mem_eff_attention_131_0 = reinterpret_cast<decltype(mem_eff_attention_131_0)>(blob_ptr + 0);
    reshape_134_0 = reinterpret_cast<decltype(reshape_134_0)>(blob_ptr + 50201856);
    layernorm_135_0 = reinterpret_cast<decltype(layernorm_135_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_136_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_136_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_137_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_137_0)>(blob_ptr + 63638784);
    reshape_139_0 = reinterpret_cast<decltype(reshape_139_0)>(blob_ptr + 40310784);
    reshape_141_0 = reinterpret_cast<decltype(reshape_141_0)>(blob_ptr + 0);
    split_142_0 = reinterpret_cast<decltype(split_142_0)>(blob_ptr + 90512640);
    split_142_1 = reinterpret_cast<decltype(split_142_1)>(blob_ptr + 77075712);
    split_142_2 = reinterpret_cast<decltype(split_142_2)>(blob_ptr + 40310784);
    mem_eff_attention_146_0 = reinterpret_cast<decltype(mem_eff_attention_146_0)>(blob_ptr + 0);
    reshape_149_0 = reinterpret_cast<decltype(reshape_149_0)>(blob_ptr + 50201856);
    layernorm_150_0 = reinterpret_cast<decltype(layernorm_150_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_151_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_151_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_152_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_152_0)>(blob_ptr + 63638784);
    reshape_154_0 = reinterpret_cast<decltype(reshape_154_0)>(blob_ptr + 40310784);
    reshape_156_0 = reinterpret_cast<decltype(reshape_156_0)>(blob_ptr + 0);
    split_157_0 = reinterpret_cast<decltype(split_157_0)>(blob_ptr + 90512640);
    split_157_1 = reinterpret_cast<decltype(split_157_1)>(blob_ptr + 40310784);
    split_157_2 = reinterpret_cast<decltype(split_157_2)>(blob_ptr + 77075712);
    mem_eff_attention_161_0 = reinterpret_cast<decltype(mem_eff_attention_161_0)>(blob_ptr + 0);
    reshape_164_0 = reinterpret_cast<decltype(reshape_164_0)>(blob_ptr + 50201856);
    layernorm_165_0 = reinterpret_cast<decltype(layernorm_165_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_166_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_166_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_167_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_167_0)>(blob_ptr + 63638784);
    reshape_169_0 = reinterpret_cast<decltype(reshape_169_0)>(blob_ptr + 40310784);
    reshape_171_0 = reinterpret_cast<decltype(reshape_171_0)>(blob_ptr + 0);
    split_172_0 = reinterpret_cast<decltype(split_172_0)>(blob_ptr + 40310784);
    split_172_1 = reinterpret_cast<decltype(split_172_1)>(blob_ptr + 77075712);
    split_172_2 = reinterpret_cast<decltype(split_172_2)>(blob_ptr + 90512640);
    mem_eff_attention_176_0 = reinterpret_cast<decltype(mem_eff_attention_176_0)>(blob_ptr + 0);
    reshape_179_0 = reinterpret_cast<decltype(reshape_179_0)>(blob_ptr + 50201856);
    layernorm_180_0 = reinterpret_cast<decltype(layernorm_180_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_181_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_181_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_182_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_182_0)>(blob_ptr + 63638784);
    reshape_184_0 = reinterpret_cast<decltype(reshape_184_0)>(blob_ptr + 40310784);
    reshape_186_0 = reinterpret_cast<decltype(reshape_186_0)>(blob_ptr + 0);
    split_187_0 = reinterpret_cast<decltype(split_187_0)>(blob_ptr + 77075712);
    split_187_1 = reinterpret_cast<decltype(split_187_1)>(blob_ptr + 90512640);
    split_187_2 = reinterpret_cast<decltype(split_187_2)>(blob_ptr + 40310784);
    mem_eff_attention_191_0 = reinterpret_cast<decltype(mem_eff_attention_191_0)>(blob_ptr + 0);
    reshape_194_0 = reinterpret_cast<decltype(reshape_194_0)>(blob_ptr + 50201856);
    layernorm_195_0 = reinterpret_cast<decltype(layernorm_195_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_196_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_196_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_197_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_197_0)>(blob_ptr + 63638784);
    reshape_199_0 = reinterpret_cast<decltype(reshape_199_0)>(blob_ptr + 40310784);
    reshape_201_0 = reinterpret_cast<decltype(reshape_201_0)>(blob_ptr + 0);
    split_202_0 = reinterpret_cast<decltype(split_202_0)>(blob_ptr + 90512640);
    split_202_1 = reinterpret_cast<decltype(split_202_1)>(blob_ptr + 77075712);
    split_202_2 = reinterpret_cast<decltype(split_202_2)>(blob_ptr + 40310784);
    mem_eff_attention_206_0 = reinterpret_cast<decltype(mem_eff_attention_206_0)>(blob_ptr + 0);
    reshape_209_0 = reinterpret_cast<decltype(reshape_209_0)>(blob_ptr + 50201856);
    layernorm_210_0 = reinterpret_cast<decltype(layernorm_210_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_211_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_211_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_212_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_212_0)>(blob_ptr + 63638784);
    reshape_214_0 = reinterpret_cast<decltype(reshape_214_0)>(blob_ptr + 40310784);
    reshape_216_0 = reinterpret_cast<decltype(reshape_216_0)>(blob_ptr + 0);
    split_217_0 = reinterpret_cast<decltype(split_217_0)>(blob_ptr + 77075712);
    split_217_1 = reinterpret_cast<decltype(split_217_1)>(blob_ptr + 90512640);
    split_217_2 = reinterpret_cast<decltype(split_217_2)>(blob_ptr + 40310784);
    mem_eff_attention_221_0 = reinterpret_cast<decltype(mem_eff_attention_221_0)>(blob_ptr + 0);
    reshape_224_0 = reinterpret_cast<decltype(reshape_224_0)>(blob_ptr + 50201856);
    layernorm_225_0 = reinterpret_cast<decltype(layernorm_225_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_226_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_226_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_227_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_227_0)>(blob_ptr + 63638784);
    reshape_229_0 = reinterpret_cast<decltype(reshape_229_0)>(blob_ptr + 40310784);
    reshape_231_0 = reinterpret_cast<decltype(reshape_231_0)>(blob_ptr + 0);
    split_232_0 = reinterpret_cast<decltype(split_232_0)>(blob_ptr + 40310784);
    split_232_1 = reinterpret_cast<decltype(split_232_1)>(blob_ptr + 77075712);
    split_232_2 = reinterpret_cast<decltype(split_232_2)>(blob_ptr + 90512640);
    mem_eff_attention_236_0 = reinterpret_cast<decltype(mem_eff_attention_236_0)>(blob_ptr + 0);
    reshape_239_0 = reinterpret_cast<decltype(reshape_239_0)>(blob_ptr + 50201856);
    layernorm_240_0 = reinterpret_cast<decltype(layernorm_240_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_241_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_241_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_242_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_242_0)>(blob_ptr + 63638784);
    reshape_244_0 = reinterpret_cast<decltype(reshape_244_0)>(blob_ptr + 40310784);
    reshape_246_0 = reinterpret_cast<decltype(reshape_246_0)>(blob_ptr + 0);
    split_247_0 = reinterpret_cast<decltype(split_247_0)>(blob_ptr + 77075712);
    split_247_1 = reinterpret_cast<decltype(split_247_1)>(blob_ptr + 40310784);
    split_247_2 = reinterpret_cast<decltype(split_247_2)>(blob_ptr + 90512640);
    mem_eff_attention_251_0 = reinterpret_cast<decltype(mem_eff_attention_251_0)>(blob_ptr + 0);
    reshape_254_0 = reinterpret_cast<decltype(reshape_254_0)>(blob_ptr + 50201856);
    layernorm_255_0 = reinterpret_cast<decltype(layernorm_255_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_256_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_256_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_257_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_257_0)>(blob_ptr + 63638784);
    reshape_259_0 = reinterpret_cast<decltype(reshape_259_0)>(blob_ptr + 40310784);
    reshape_261_0 = reinterpret_cast<decltype(reshape_261_0)>(blob_ptr + 0);
    split_262_0 = reinterpret_cast<decltype(split_262_0)>(blob_ptr + 40310784);
    split_262_1 = reinterpret_cast<decltype(split_262_1)>(blob_ptr + 77075712);
    split_262_2 = reinterpret_cast<decltype(split_262_2)>(blob_ptr + 90512640);
    mem_eff_attention_266_0 = reinterpret_cast<decltype(mem_eff_attention_266_0)>(blob_ptr + 0);
    reshape_269_0 = reinterpret_cast<decltype(reshape_269_0)>(blob_ptr + 50201856);
    layernorm_270_0 = reinterpret_cast<decltype(layernorm_270_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_271_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_271_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_272_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_272_0)>(blob_ptr + 63638784);
    reshape_274_0 = reinterpret_cast<decltype(reshape_274_0)>(blob_ptr + 40310784);
    reshape_276_0 = reinterpret_cast<decltype(reshape_276_0)>(blob_ptr + 0);
    split_277_0 = reinterpret_cast<decltype(split_277_0)>(blob_ptr + 77075712);
    split_277_1 = reinterpret_cast<decltype(split_277_1)>(blob_ptr + 40310784);
    split_277_2 = reinterpret_cast<decltype(split_277_2)>(blob_ptr + 90512640);
    mem_eff_attention_281_0 = reinterpret_cast<decltype(mem_eff_attention_281_0)>(blob_ptr + 0);
    reshape_284_0 = reinterpret_cast<decltype(reshape_284_0)>(blob_ptr + 50201856);
    layernorm_285_0 = reinterpret_cast<decltype(layernorm_285_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_286_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_286_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_287_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_287_0)>(blob_ptr + 63638784);
    reshape_289_0 = reinterpret_cast<decltype(reshape_289_0)>(blob_ptr + 40310784);
    reshape_291_0 = reinterpret_cast<decltype(reshape_291_0)>(blob_ptr + 0);
    split_292_0 = reinterpret_cast<decltype(split_292_0)>(blob_ptr + 40310784);
    split_292_1 = reinterpret_cast<decltype(split_292_1)>(blob_ptr + 77075712);
    split_292_2 = reinterpret_cast<decltype(split_292_2)>(blob_ptr + 90512640);
    mem_eff_attention_296_0 = reinterpret_cast<decltype(mem_eff_attention_296_0)>(blob_ptr + 0);
    reshape_299_0 = reinterpret_cast<decltype(reshape_299_0)>(blob_ptr + 50201856);
    layernorm_300_0 = reinterpret_cast<decltype(layernorm_300_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_301_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_301_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_302_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_302_0)>(blob_ptr + 63638784);
    reshape_304_0 = reinterpret_cast<decltype(reshape_304_0)>(blob_ptr + 40310784);
    reshape_306_0 = reinterpret_cast<decltype(reshape_306_0)>(blob_ptr + 0);
    split_307_0 = reinterpret_cast<decltype(split_307_0)>(blob_ptr + 90512640);
    split_307_1 = reinterpret_cast<decltype(split_307_1)>(blob_ptr + 77075712);
    split_307_2 = reinterpret_cast<decltype(split_307_2)>(blob_ptr + 40310784);
    mem_eff_attention_311_0 = reinterpret_cast<decltype(mem_eff_attention_311_0)>(blob_ptr + 0);
    reshape_314_0 = reinterpret_cast<decltype(reshape_314_0)>(blob_ptr + 50201856);
    layernorm_315_0 = reinterpret_cast<decltype(layernorm_315_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_316_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_316_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_317_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_317_0)>(blob_ptr + 63638784);
    reshape_319_0 = reinterpret_cast<decltype(reshape_319_0)>(blob_ptr + 40310784);
    reshape_321_0 = reinterpret_cast<decltype(reshape_321_0)>(blob_ptr + 0);
    split_322_0 = reinterpret_cast<decltype(split_322_0)>(blob_ptr + 90512640);
    split_322_1 = reinterpret_cast<decltype(split_322_1)>(blob_ptr + 40310784);
    split_322_2 = reinterpret_cast<decltype(split_322_2)>(blob_ptr + 77075712);
    mem_eff_attention_326_0 = reinterpret_cast<decltype(mem_eff_attention_326_0)>(blob_ptr + 0);
    reshape_329_0 = reinterpret_cast<decltype(reshape_329_0)>(blob_ptr + 50201856);
    layernorm_330_0 = reinterpret_cast<decltype(layernorm_330_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_331_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_331_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_332_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_332_0)>(blob_ptr + 63638784);
    reshape_334_0 = reinterpret_cast<decltype(reshape_334_0)>(blob_ptr + 40310784);
    reshape_336_0 = reinterpret_cast<decltype(reshape_336_0)>(blob_ptr + 0);
    split_337_0 = reinterpret_cast<decltype(split_337_0)>(blob_ptr + 40310784);
    split_337_1 = reinterpret_cast<decltype(split_337_1)>(blob_ptr + 77075712);
    split_337_2 = reinterpret_cast<decltype(split_337_2)>(blob_ptr + 90512640);
    mem_eff_attention_341_0 = reinterpret_cast<decltype(mem_eff_attention_341_0)>(blob_ptr + 0);
    reshape_344_0 = reinterpret_cast<decltype(reshape_344_0)>(blob_ptr + 50201856);
    layernorm_345_0 = reinterpret_cast<decltype(layernorm_345_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_346_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_346_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_347_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_347_0)>(blob_ptr + 63638784);
    reshape_349_0 = reinterpret_cast<decltype(reshape_349_0)>(blob_ptr + 40310784);
    reshape_351_0 = reinterpret_cast<decltype(reshape_351_0)>(blob_ptr + 0);
    split_352_0 = reinterpret_cast<decltype(split_352_0)>(blob_ptr + 90512640);
    split_352_1 = reinterpret_cast<decltype(split_352_1)>(blob_ptr + 40310784);
    split_352_2 = reinterpret_cast<decltype(split_352_2)>(blob_ptr + 77075712);
    mem_eff_attention_356_0 = reinterpret_cast<decltype(mem_eff_attention_356_0)>(blob_ptr + 0);
    reshape_359_0 = reinterpret_cast<decltype(reshape_359_0)>(blob_ptr + 50201856);
    layernorm_360_0 = reinterpret_cast<decltype(layernorm_360_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_361_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_361_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_362_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_362_0)>(blob_ptr + 63638784);
    reshape_364_0 = reinterpret_cast<decltype(reshape_364_0)>(blob_ptr + 40310784);
    reshape_366_0 = reinterpret_cast<decltype(reshape_366_0)>(blob_ptr + 0);
    split_367_0 = reinterpret_cast<decltype(split_367_0)>(blob_ptr + 40310784);
    split_367_1 = reinterpret_cast<decltype(split_367_1)>(blob_ptr + 90512640);
    split_367_2 = reinterpret_cast<decltype(split_367_2)>(blob_ptr + 77075712);
    mem_eff_attention_371_0 = reinterpret_cast<decltype(mem_eff_attention_371_0)>(blob_ptr + 0);
    reshape_374_0 = reinterpret_cast<decltype(reshape_374_0)>(blob_ptr + 50201856);
    layernorm_375_0 = reinterpret_cast<decltype(layernorm_375_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_376_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_376_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_377_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_377_0)>(blob_ptr + 63638784);
    reshape_379_0 = reinterpret_cast<decltype(reshape_379_0)>(blob_ptr + 40310784);
    reshape_381_0 = reinterpret_cast<decltype(reshape_381_0)>(blob_ptr + 0);
    split_382_0 = reinterpret_cast<decltype(split_382_0)>(blob_ptr + 77075712);
    split_382_1 = reinterpret_cast<decltype(split_382_1)>(blob_ptr + 40310784);
    split_382_2 = reinterpret_cast<decltype(split_382_2)>(blob_ptr + 90512640);
    mem_eff_attention_386_0 = reinterpret_cast<decltype(mem_eff_attention_386_0)>(blob_ptr + 0);
    reshape_389_0 = reinterpret_cast<decltype(reshape_389_0)>(blob_ptr + 50201856);
    layernorm_390_0 = reinterpret_cast<decltype(layernorm_390_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_391_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_391_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_392_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_392_0)>(blob_ptr + 63638784);
    reshape_394_0 = reinterpret_cast<decltype(reshape_394_0)>(blob_ptr + 40310784);
    reshape_396_0 = reinterpret_cast<decltype(reshape_396_0)>(blob_ptr + 0);
    split_397_0 = reinterpret_cast<decltype(split_397_0)>(blob_ptr + 90512640);
    split_397_1 = reinterpret_cast<decltype(split_397_1)>(blob_ptr + 40310784);
    split_397_2 = reinterpret_cast<decltype(split_397_2)>(blob_ptr + 77075712);
    mem_eff_attention_401_0 = reinterpret_cast<decltype(mem_eff_attention_401_0)>(blob_ptr + 0);
    reshape_404_0 = reinterpret_cast<decltype(reshape_404_0)>(blob_ptr + 50201856);
    layernorm_405_0 = reinterpret_cast<decltype(layernorm_405_0)>(blob_ptr + 63638784);
    gemm_rcr_bias_gelu_406_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_406_0)>(blob_ptr + 0);
    gemm_rcr_bias_add_407_0 = reinterpret_cast<decltype(gemm_rcr_bias_add_407_0)>(blob_ptr + 63638784);
    layernorm_408_0 = reinterpret_cast<decltype(layernorm_408_0)>(blob_ptr + 26873856);
    reshape_410_0 = reinterpret_cast<decltype(reshape_410_0)>(blob_ptr + 0);
    permute_411_0 = reinterpret_cast<decltype(permute_411_0)>(blob_ptr + 26873856);
    split_412_0 = reinterpret_cast<decltype(split_412_0)>(blob_ptr + 13436928);
    split_412_1 = reinterpret_cast<decltype(split_412_1)>(blob_ptr + 0);
    reshape_417_0 = reinterpret_cast<decltype(reshape_417_0)>(constants + 2658816);
    mem_eff_attention_418_0 = reinterpret_cast<decltype(mem_eff_attention_418_0)>(blob_ptr + 26873856);
    gemm_rcr_bias_421_0 = reinterpret_cast<decltype(gemm_rcr_bias_421_0)>(blob_ptr + 68864);
    layernorm_422_0 = reinterpret_cast<decltype(layernorm_422_0)>(blob_ptr + 87296);
    gemm_rcr_bias_gelu_423_0 = reinterpret_cast<decltype(gemm_rcr_bias_gelu_423_0)>(blob_ptr + 0);
    
         params_[2].shape_ptrs = {ParamDim(1, 1, &pool_probe_dim_0), ParamDim(1, 1, &pool_probe_dim_1), ParamDim(1152, 1152, &pool_probe_dim_2)};
     params_[3].shape_ptrs = {ParamDim(1152, 1152, &pool_q_weight_dim_0), ParamDim(1152, 1152, &pool_q_weight_dim_1)};
     params_[4].shape_ptrs = {ParamDim(1152, 1152, &pool_q_bias_dim_0)};
     params_[0].shape_ptrs = {ParamDim(8, 8, &input0_dim_0), ParamDim(384, 384, &input0_dim_1), ParamDim(384, 384, &input0_dim_2), ParamDim(3, 3, &input0_dim_3)};
     params_[5].shape_ptrs = {ParamDim(1152, 1152, &patch_embed_proj_weight_dim_0), ParamDim(14, 14, &patch_embed_proj_weight_dim_1), ParamDim(14, 14, &patch_embed_proj_weight_dim_2), ParamDim(3, 3, &patch_embed_proj_weight_dim_3)};
     params_[6].shape_ptrs = {ParamDim(1152, 1152, &patch_embed_proj_bias_dim_0)};
     params_[7].shape_ptrs = {ParamDim(1, 1, &pos_emb_pos_emb_dim_0), ParamDim(729, 729, &pos_emb_pos_emb_dim_1), ParamDim(1152, 1152, &pos_emb_pos_emb_dim_2)};
     params_[8].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_0_ln1_weight_dim_0)};
     params_[9].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_0_ln1_bias_dim_0)};
     params_[10].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_0_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_0_mha_qkv_weight_dim_1)};
     params_[11].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_0_mha_qkv_bias_dim_0)};
     params_[12].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_0_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_0_mha_proj_weight_dim_1)};
     params_[13].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_0_mha_proj_bias_dim_0)};
     params_[14].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_0_ln2_weight_dim_0)};
     params_[15].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_0_ln2_bias_dim_0)};
     params_[16].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_0_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_0_mlp_fc1_weight_dim_1)};
     params_[17].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_0_mlp_fc1_bias_dim_0)};
     params_[18].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_0_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_0_mlp_fc2_weight_dim_1)};
     params_[19].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_0_mlp_fc2_bias_dim_0)};
     params_[20].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_1_ln1_weight_dim_0)};
     params_[21].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_1_ln1_bias_dim_0)};
     params_[22].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_1_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_1_mha_qkv_weight_dim_1)};
     params_[23].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_1_mha_qkv_bias_dim_0)};
     params_[24].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_1_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_1_mha_proj_weight_dim_1)};
     params_[25].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_1_mha_proj_bias_dim_0)};
     params_[26].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_1_ln2_weight_dim_0)};
     params_[27].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_1_ln2_bias_dim_0)};
     params_[28].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_1_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_1_mlp_fc1_weight_dim_1)};
     params_[29].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_1_mlp_fc1_bias_dim_0)};
     params_[30].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_1_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_1_mlp_fc2_weight_dim_1)};
     params_[31].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_1_mlp_fc2_bias_dim_0)};
     params_[32].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_2_ln1_weight_dim_0)};
     params_[33].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_2_ln1_bias_dim_0)};
     params_[34].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_2_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_2_mha_qkv_weight_dim_1)};
     params_[35].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_2_mha_qkv_bias_dim_0)};
     params_[36].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_2_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_2_mha_proj_weight_dim_1)};
     params_[37].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_2_mha_proj_bias_dim_0)};
     params_[38].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_2_ln2_weight_dim_0)};
     params_[39].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_2_ln2_bias_dim_0)};
     params_[40].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_2_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_2_mlp_fc1_weight_dim_1)};
     params_[41].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_2_mlp_fc1_bias_dim_0)};
     params_[42].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_2_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_2_mlp_fc2_weight_dim_1)};
     params_[43].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_2_mlp_fc2_bias_dim_0)};
     params_[44].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_3_ln1_weight_dim_0)};
     params_[45].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_3_ln1_bias_dim_0)};
     params_[46].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_3_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_3_mha_qkv_weight_dim_1)};
     params_[47].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_3_mha_qkv_bias_dim_0)};
     params_[48].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_3_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_3_mha_proj_weight_dim_1)};
     params_[49].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_3_mha_proj_bias_dim_0)};
     params_[50].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_3_ln2_weight_dim_0)};
     params_[51].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_3_ln2_bias_dim_0)};
     params_[52].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_3_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_3_mlp_fc1_weight_dim_1)};
     params_[53].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_3_mlp_fc1_bias_dim_0)};
     params_[54].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_3_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_3_mlp_fc2_weight_dim_1)};
     params_[55].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_3_mlp_fc2_bias_dim_0)};
     params_[56].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_4_ln1_weight_dim_0)};
     params_[57].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_4_ln1_bias_dim_0)};
     params_[58].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_4_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_4_mha_qkv_weight_dim_1)};
     params_[59].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_4_mha_qkv_bias_dim_0)};
     params_[60].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_4_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_4_mha_proj_weight_dim_1)};
     params_[61].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_4_mha_proj_bias_dim_0)};
     params_[62].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_4_ln2_weight_dim_0)};
     params_[63].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_4_ln2_bias_dim_0)};
     params_[64].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_4_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_4_mlp_fc1_weight_dim_1)};
     params_[65].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_4_mlp_fc1_bias_dim_0)};
     params_[66].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_4_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_4_mlp_fc2_weight_dim_1)};
     params_[67].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_4_mlp_fc2_bias_dim_0)};
     params_[68].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_5_ln1_weight_dim_0)};
     params_[69].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_5_ln1_bias_dim_0)};
     params_[70].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_5_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_5_mha_qkv_weight_dim_1)};
     params_[71].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_5_mha_qkv_bias_dim_0)};
     params_[72].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_5_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_5_mha_proj_weight_dim_1)};
     params_[73].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_5_mha_proj_bias_dim_0)};
     params_[74].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_5_ln2_weight_dim_0)};
     params_[75].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_5_ln2_bias_dim_0)};
     params_[76].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_5_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_5_mlp_fc1_weight_dim_1)};
     params_[77].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_5_mlp_fc1_bias_dim_0)};
     params_[78].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_5_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_5_mlp_fc2_weight_dim_1)};
     params_[79].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_5_mlp_fc2_bias_dim_0)};
     params_[80].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_6_ln1_weight_dim_0)};
     params_[81].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_6_ln1_bias_dim_0)};
     params_[82].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_6_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_6_mha_qkv_weight_dim_1)};
     params_[83].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_6_mha_qkv_bias_dim_0)};
     params_[84].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_6_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_6_mha_proj_weight_dim_1)};
     params_[85].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_6_mha_proj_bias_dim_0)};
     params_[86].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_6_ln2_weight_dim_0)};
     params_[87].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_6_ln2_bias_dim_0)};
     params_[88].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_6_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_6_mlp_fc1_weight_dim_1)};
     params_[89].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_6_mlp_fc1_bias_dim_0)};
     params_[90].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_6_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_6_mlp_fc2_weight_dim_1)};
     params_[91].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_6_mlp_fc2_bias_dim_0)};
     params_[92].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_7_ln1_weight_dim_0)};
     params_[93].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_7_ln1_bias_dim_0)};
     params_[94].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_7_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_7_mha_qkv_weight_dim_1)};
     params_[95].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_7_mha_qkv_bias_dim_0)};
     params_[96].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_7_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_7_mha_proj_weight_dim_1)};
     params_[97].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_7_mha_proj_bias_dim_0)};
     params_[98].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_7_ln2_weight_dim_0)};
     params_[99].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_7_ln2_bias_dim_0)};
     params_[100].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_7_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_7_mlp_fc1_weight_dim_1)};
     params_[101].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_7_mlp_fc1_bias_dim_0)};
     params_[102].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_7_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_7_mlp_fc2_weight_dim_1)};
     params_[103].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_7_mlp_fc2_bias_dim_0)};
     params_[104].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_8_ln1_weight_dim_0)};
     params_[105].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_8_ln1_bias_dim_0)};
     params_[106].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_8_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_8_mha_qkv_weight_dim_1)};
     params_[107].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_8_mha_qkv_bias_dim_0)};
     params_[108].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_8_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_8_mha_proj_weight_dim_1)};
     params_[109].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_8_mha_proj_bias_dim_0)};
     params_[110].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_8_ln2_weight_dim_0)};
     params_[111].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_8_ln2_bias_dim_0)};
     params_[112].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_8_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_8_mlp_fc1_weight_dim_1)};
     params_[113].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_8_mlp_fc1_bias_dim_0)};
     params_[114].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_8_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_8_mlp_fc2_weight_dim_1)};
     params_[115].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_8_mlp_fc2_bias_dim_0)};
     params_[116].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_9_ln1_weight_dim_0)};
     params_[117].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_9_ln1_bias_dim_0)};
     params_[118].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_9_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_9_mha_qkv_weight_dim_1)};
     params_[119].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_9_mha_qkv_bias_dim_0)};
     params_[120].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_9_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_9_mha_proj_weight_dim_1)};
     params_[121].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_9_mha_proj_bias_dim_0)};
     params_[122].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_9_ln2_weight_dim_0)};
     params_[123].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_9_ln2_bias_dim_0)};
     params_[124].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_9_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_9_mlp_fc1_weight_dim_1)};
     params_[125].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_9_mlp_fc1_bias_dim_0)};
     params_[126].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_9_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_9_mlp_fc2_weight_dim_1)};
     params_[127].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_9_mlp_fc2_bias_dim_0)};
     params_[128].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_10_ln1_weight_dim_0)};
     params_[129].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_10_ln1_bias_dim_0)};
     params_[130].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_10_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_10_mha_qkv_weight_dim_1)};
     params_[131].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_10_mha_qkv_bias_dim_0)};
     params_[132].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_10_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_10_mha_proj_weight_dim_1)};
     params_[133].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_10_mha_proj_bias_dim_0)};
     params_[134].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_10_ln2_weight_dim_0)};
     params_[135].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_10_ln2_bias_dim_0)};
     params_[136].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_10_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_10_mlp_fc1_weight_dim_1)};
     params_[137].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_10_mlp_fc1_bias_dim_0)};
     params_[138].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_10_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_10_mlp_fc2_weight_dim_1)};
     params_[139].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_10_mlp_fc2_bias_dim_0)};
     params_[140].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_11_ln1_weight_dim_0)};
     params_[141].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_11_ln1_bias_dim_0)};
     params_[142].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_11_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_11_mha_qkv_weight_dim_1)};
     params_[143].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_11_mha_qkv_bias_dim_0)};
     params_[144].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_11_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_11_mha_proj_weight_dim_1)};
     params_[145].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_11_mha_proj_bias_dim_0)};
     params_[146].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_11_ln2_weight_dim_0)};
     params_[147].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_11_ln2_bias_dim_0)};
     params_[148].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_11_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_11_mlp_fc1_weight_dim_1)};
     params_[149].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_11_mlp_fc1_bias_dim_0)};
     params_[150].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_11_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_11_mlp_fc2_weight_dim_1)};
     params_[151].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_11_mlp_fc2_bias_dim_0)};
     params_[152].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_12_ln1_weight_dim_0)};
     params_[153].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_12_ln1_bias_dim_0)};
     params_[154].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_12_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_12_mha_qkv_weight_dim_1)};
     params_[155].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_12_mha_qkv_bias_dim_0)};
     params_[156].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_12_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_12_mha_proj_weight_dim_1)};
     params_[157].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_12_mha_proj_bias_dim_0)};
     params_[158].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_12_ln2_weight_dim_0)};
     params_[159].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_12_ln2_bias_dim_0)};
     params_[160].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_12_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_12_mlp_fc1_weight_dim_1)};
     params_[161].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_12_mlp_fc1_bias_dim_0)};
     params_[162].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_12_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_12_mlp_fc2_weight_dim_1)};
     params_[163].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_12_mlp_fc2_bias_dim_0)};
     params_[164].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_13_ln1_weight_dim_0)};
     params_[165].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_13_ln1_bias_dim_0)};
     params_[166].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_13_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_13_mha_qkv_weight_dim_1)};
     params_[167].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_13_mha_qkv_bias_dim_0)};
     params_[168].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_13_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_13_mha_proj_weight_dim_1)};
     params_[169].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_13_mha_proj_bias_dim_0)};
     params_[170].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_13_ln2_weight_dim_0)};
     params_[171].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_13_ln2_bias_dim_0)};
     params_[172].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_13_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_13_mlp_fc1_weight_dim_1)};
     params_[173].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_13_mlp_fc1_bias_dim_0)};
     params_[174].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_13_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_13_mlp_fc2_weight_dim_1)};
     params_[175].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_13_mlp_fc2_bias_dim_0)};
     params_[176].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_14_ln1_weight_dim_0)};
     params_[177].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_14_ln1_bias_dim_0)};
     params_[178].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_14_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_14_mha_qkv_weight_dim_1)};
     params_[179].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_14_mha_qkv_bias_dim_0)};
     params_[180].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_14_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_14_mha_proj_weight_dim_1)};
     params_[181].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_14_mha_proj_bias_dim_0)};
     params_[182].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_14_ln2_weight_dim_0)};
     params_[183].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_14_ln2_bias_dim_0)};
     params_[184].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_14_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_14_mlp_fc1_weight_dim_1)};
     params_[185].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_14_mlp_fc1_bias_dim_0)};
     params_[186].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_14_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_14_mlp_fc2_weight_dim_1)};
     params_[187].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_14_mlp_fc2_bias_dim_0)};
     params_[188].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_15_ln1_weight_dim_0)};
     params_[189].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_15_ln1_bias_dim_0)};
     params_[190].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_15_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_15_mha_qkv_weight_dim_1)};
     params_[191].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_15_mha_qkv_bias_dim_0)};
     params_[192].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_15_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_15_mha_proj_weight_dim_1)};
     params_[193].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_15_mha_proj_bias_dim_0)};
     params_[194].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_15_ln2_weight_dim_0)};
     params_[195].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_15_ln2_bias_dim_0)};
     params_[196].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_15_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_15_mlp_fc1_weight_dim_1)};
     params_[197].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_15_mlp_fc1_bias_dim_0)};
     params_[198].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_15_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_15_mlp_fc2_weight_dim_1)};
     params_[199].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_15_mlp_fc2_bias_dim_0)};
     params_[200].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_16_ln1_weight_dim_0)};
     params_[201].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_16_ln1_bias_dim_0)};
     params_[202].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_16_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_16_mha_qkv_weight_dim_1)};
     params_[203].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_16_mha_qkv_bias_dim_0)};
     params_[204].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_16_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_16_mha_proj_weight_dim_1)};
     params_[205].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_16_mha_proj_bias_dim_0)};
     params_[206].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_16_ln2_weight_dim_0)};
     params_[207].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_16_ln2_bias_dim_0)};
     params_[208].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_16_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_16_mlp_fc1_weight_dim_1)};
     params_[209].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_16_mlp_fc1_bias_dim_0)};
     params_[210].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_16_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_16_mlp_fc2_weight_dim_1)};
     params_[211].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_16_mlp_fc2_bias_dim_0)};
     params_[212].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_17_ln1_weight_dim_0)};
     params_[213].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_17_ln1_bias_dim_0)};
     params_[214].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_17_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_17_mha_qkv_weight_dim_1)};
     params_[215].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_17_mha_qkv_bias_dim_0)};
     params_[216].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_17_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_17_mha_proj_weight_dim_1)};
     params_[217].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_17_mha_proj_bias_dim_0)};
     params_[218].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_17_ln2_weight_dim_0)};
     params_[219].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_17_ln2_bias_dim_0)};
     params_[220].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_17_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_17_mlp_fc1_weight_dim_1)};
     params_[221].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_17_mlp_fc1_bias_dim_0)};
     params_[222].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_17_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_17_mlp_fc2_weight_dim_1)};
     params_[223].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_17_mlp_fc2_bias_dim_0)};
     params_[224].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_18_ln1_weight_dim_0)};
     params_[225].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_18_ln1_bias_dim_0)};
     params_[226].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_18_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_18_mha_qkv_weight_dim_1)};
     params_[227].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_18_mha_qkv_bias_dim_0)};
     params_[228].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_18_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_18_mha_proj_weight_dim_1)};
     params_[229].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_18_mha_proj_bias_dim_0)};
     params_[230].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_18_ln2_weight_dim_0)};
     params_[231].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_18_ln2_bias_dim_0)};
     params_[232].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_18_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_18_mlp_fc1_weight_dim_1)};
     params_[233].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_18_mlp_fc1_bias_dim_0)};
     params_[234].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_18_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_18_mlp_fc2_weight_dim_1)};
     params_[235].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_18_mlp_fc2_bias_dim_0)};
     params_[236].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_19_ln1_weight_dim_0)};
     params_[237].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_19_ln1_bias_dim_0)};
     params_[238].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_19_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_19_mha_qkv_weight_dim_1)};
     params_[239].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_19_mha_qkv_bias_dim_0)};
     params_[240].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_19_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_19_mha_proj_weight_dim_1)};
     params_[241].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_19_mha_proj_bias_dim_0)};
     params_[242].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_19_ln2_weight_dim_0)};
     params_[243].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_19_ln2_bias_dim_0)};
     params_[244].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_19_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_19_mlp_fc1_weight_dim_1)};
     params_[245].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_19_mlp_fc1_bias_dim_0)};
     params_[246].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_19_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_19_mlp_fc2_weight_dim_1)};
     params_[247].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_19_mlp_fc2_bias_dim_0)};
     params_[248].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_20_ln1_weight_dim_0)};
     params_[249].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_20_ln1_bias_dim_0)};
     params_[250].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_20_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_20_mha_qkv_weight_dim_1)};
     params_[251].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_20_mha_qkv_bias_dim_0)};
     params_[252].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_20_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_20_mha_proj_weight_dim_1)};
     params_[253].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_20_mha_proj_bias_dim_0)};
     params_[254].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_20_ln2_weight_dim_0)};
     params_[255].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_20_ln2_bias_dim_0)};
     params_[256].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_20_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_20_mlp_fc1_weight_dim_1)};
     params_[257].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_20_mlp_fc1_bias_dim_0)};
     params_[258].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_20_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_20_mlp_fc2_weight_dim_1)};
     params_[259].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_20_mlp_fc2_bias_dim_0)};
     params_[260].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_21_ln1_weight_dim_0)};
     params_[261].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_21_ln1_bias_dim_0)};
     params_[262].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_21_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_21_mha_qkv_weight_dim_1)};
     params_[263].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_21_mha_qkv_bias_dim_0)};
     params_[264].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_21_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_21_mha_proj_weight_dim_1)};
     params_[265].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_21_mha_proj_bias_dim_0)};
     params_[266].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_21_ln2_weight_dim_0)};
     params_[267].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_21_ln2_bias_dim_0)};
     params_[268].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_21_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_21_mlp_fc1_weight_dim_1)};
     params_[269].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_21_mlp_fc1_bias_dim_0)};
     params_[270].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_21_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_21_mlp_fc2_weight_dim_1)};
     params_[271].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_21_mlp_fc2_bias_dim_0)};
     params_[272].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_22_ln1_weight_dim_0)};
     params_[273].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_22_ln1_bias_dim_0)};
     params_[274].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_22_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_22_mha_qkv_weight_dim_1)};
     params_[275].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_22_mha_qkv_bias_dim_0)};
     params_[276].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_22_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_22_mha_proj_weight_dim_1)};
     params_[277].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_22_mha_proj_bias_dim_0)};
     params_[278].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_22_ln2_weight_dim_0)};
     params_[279].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_22_ln2_bias_dim_0)};
     params_[280].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_22_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_22_mlp_fc1_weight_dim_1)};
     params_[281].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_22_mlp_fc1_bias_dim_0)};
     params_[282].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_22_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_22_mlp_fc2_weight_dim_1)};
     params_[283].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_22_mlp_fc2_bias_dim_0)};
     params_[284].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_23_ln1_weight_dim_0)};
     params_[285].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_23_ln1_bias_dim_0)};
     params_[286].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_23_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_23_mha_qkv_weight_dim_1)};
     params_[287].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_23_mha_qkv_bias_dim_0)};
     params_[288].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_23_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_23_mha_proj_weight_dim_1)};
     params_[289].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_23_mha_proj_bias_dim_0)};
     params_[290].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_23_ln2_weight_dim_0)};
     params_[291].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_23_ln2_bias_dim_0)};
     params_[292].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_23_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_23_mlp_fc1_weight_dim_1)};
     params_[293].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_23_mlp_fc1_bias_dim_0)};
     params_[294].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_23_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_23_mlp_fc2_weight_dim_1)};
     params_[295].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_23_mlp_fc2_bias_dim_0)};
     params_[296].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_24_ln1_weight_dim_0)};
     params_[297].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_24_ln1_bias_dim_0)};
     params_[298].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_24_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_24_mha_qkv_weight_dim_1)};
     params_[299].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_24_mha_qkv_bias_dim_0)};
     params_[300].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_24_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_24_mha_proj_weight_dim_1)};
     params_[301].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_24_mha_proj_bias_dim_0)};
     params_[302].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_24_ln2_weight_dim_0)};
     params_[303].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_24_ln2_bias_dim_0)};
     params_[304].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_24_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_24_mlp_fc1_weight_dim_1)};
     params_[305].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_24_mlp_fc1_bias_dim_0)};
     params_[306].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_24_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_24_mlp_fc2_weight_dim_1)};
     params_[307].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_24_mlp_fc2_bias_dim_0)};
     params_[308].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_25_ln1_weight_dim_0)};
     params_[309].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_25_ln1_bias_dim_0)};
     params_[310].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_25_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_25_mha_qkv_weight_dim_1)};
     params_[311].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_25_mha_qkv_bias_dim_0)};
     params_[312].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_25_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_25_mha_proj_weight_dim_1)};
     params_[313].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_25_mha_proj_bias_dim_0)};
     params_[314].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_25_ln2_weight_dim_0)};
     params_[315].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_25_ln2_bias_dim_0)};
     params_[316].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_25_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_25_mlp_fc1_weight_dim_1)};
     params_[317].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_25_mlp_fc1_bias_dim_0)};
     params_[318].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_25_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_25_mlp_fc2_weight_dim_1)};
     params_[319].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_25_mlp_fc2_bias_dim_0)};
     params_[320].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_26_ln1_weight_dim_0)};
     params_[321].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_26_ln1_bias_dim_0)};
     params_[322].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_26_mha_qkv_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_26_mha_qkv_weight_dim_1)};
     params_[323].shape_ptrs = {ParamDim(3456, 3456, &encoder_layers_26_mha_qkv_bias_dim_0)};
     params_[324].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_26_mha_proj_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_26_mha_proj_weight_dim_1)};
     params_[325].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_26_mha_proj_bias_dim_0)};
     params_[326].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_26_ln2_weight_dim_0)};
     params_[327].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_26_ln2_bias_dim_0)};
     params_[328].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_26_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &encoder_layers_26_mlp_fc1_weight_dim_1)};
     params_[329].shape_ptrs = {ParamDim(4304, 4304, &encoder_layers_26_mlp_fc1_bias_dim_0)};
     params_[330].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_26_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &encoder_layers_26_mlp_fc2_weight_dim_1)};
     params_[331].shape_ptrs = {ParamDim(1152, 1152, &encoder_layers_26_mlp_fc2_bias_dim_0)};
     params_[332].shape_ptrs = {ParamDim(1152, 1152, &encoder_ln_weight_dim_0)};
     params_[333].shape_ptrs = {ParamDim(1152, 1152, &encoder_ln_bias_dim_0)};
     params_[334].shape_ptrs = {ParamDim(2304, 2304, &pool_kv_weight_dim_0), ParamDim(1152, 1152, &pool_kv_weight_dim_1)};
     params_[335].shape_ptrs = {ParamDim(2304, 2304, &pool_kv_bias_dim_0)};
     params_[336].shape_ptrs = {ParamDim(1152, 1152, &pool_proj_weight_dim_0), ParamDim(1152, 1152, &pool_proj_weight_dim_1)};
     params_[337].shape_ptrs = {ParamDim(1152, 1152, &pool_proj_bias_dim_0)};
     params_[338].shape_ptrs = {ParamDim(1152, 1152, &pool_ln_weight_dim_0)};
     params_[339].shape_ptrs = {ParamDim(1152, 1152, &pool_ln_bias_dim_0)};
     params_[340].shape_ptrs = {ParamDim(4304, 4304, &pool_mlp_fc1_weight_dim_0), ParamDim(1152, 1152, &pool_mlp_fc1_weight_dim_1)};
     params_[341].shape_ptrs = {ParamDim(4304, 4304, &pool_mlp_fc1_bias_dim_0)};
     params_[342].shape_ptrs = {ParamDim(1152, 1152, &pool_mlp_fc2_weight_dim_0), ParamDim(4304, 4304, &pool_mlp_fc2_weight_dim_1)};
     params_[343].shape_ptrs = {ParamDim(1152, 1152, &pool_mlp_fc2_bias_dim_0)};
     params_[1].shape_ptrs = {ParamDim(8, 8, &reshape_420_0_dim_0), ParamDim(1, 1, &reshape_420_0_dim_1), ParamDim(1152, 1152, &pool_mlp_fc2_weight_dim_0)};

      
      
    }

    ~Model() {
      
      
    }

    void SetUpInputsOutputs() {
             input0 = static_cast<decltype(input0)>(params_[0].ptr);

if (input0 == nullptr) {
    throw std::runtime_error("Constant input0 was not set! Set the value with set_constant.");
}
    

if (patch_embed_proj_weight == nullptr) {
    throw std::runtime_error("Constant patch_embed_proj_weight was not set! Set the value with set_constant.");
}
    

if (patch_embed_proj_bias == nullptr) {
    throw std::runtime_error("Constant patch_embed_proj_bias was not set! Set the value with set_constant.");
}
    

if (pos_emb_pos_emb == nullptr) {
    throw std::runtime_error("Constant pos_emb_pos_emb was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_0_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_0_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_1_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_1_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_2_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_2_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_3_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_3_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_4_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_4_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_5_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_5_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_6_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_6_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_7_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_7_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_8_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_8_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_9_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_9_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_10_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_10_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_11_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_11_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_12_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_12_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_13_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_13_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_14_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_14_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_15_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_15_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_16_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_16_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_17_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_17_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_18_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_18_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_19_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_19_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_20_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_20_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_21_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_21_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_22_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_22_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_23_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_23_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_24_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_24_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_25_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_25_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_ln1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_ln1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_ln1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_ln1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_mha_qkv_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_mha_qkv_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_mha_qkv_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_mha_qkv_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_mha_proj_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_mha_proj_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_mha_proj_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_mha_proj_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_ln2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_ln2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_ln2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_ln2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (encoder_layers_26_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant encoder_layers_26_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    

if (encoder_ln_weight == nullptr) {
    throw std::runtime_error("Constant encoder_ln_weight was not set! Set the value with set_constant.");
}
    

if (encoder_ln_bias == nullptr) {
    throw std::runtime_error("Constant encoder_ln_bias was not set! Set the value with set_constant.");
}
    

if (pool_kv_weight == nullptr) {
    throw std::runtime_error("Constant pool_kv_weight was not set! Set the value with set_constant.");
}
    

if (pool_kv_bias == nullptr) {
    throw std::runtime_error("Constant pool_kv_bias was not set! Set the value with set_constant.");
}
    

if (pool_proj_weight == nullptr) {
    throw std::runtime_error("Constant pool_proj_weight was not set! Set the value with set_constant.");
}
    

if (pool_proj_bias == nullptr) {
    throw std::runtime_error("Constant pool_proj_bias was not set! Set the value with set_constant.");
}
    

if (pool_ln_weight == nullptr) {
    throw std::runtime_error("Constant pool_ln_weight was not set! Set the value with set_constant.");
}
    

if (pool_ln_bias == nullptr) {
    throw std::runtime_error("Constant pool_ln_bias was not set! Set the value with set_constant.");
}
    

if (pool_mlp_fc1_weight == nullptr) {
    throw std::runtime_error("Constant pool_mlp_fc1_weight was not set! Set the value with set_constant.");
}
    

if (pool_mlp_fc1_bias == nullptr) {
    throw std::runtime_error("Constant pool_mlp_fc1_bias was not set! Set the value with set_constant.");
}
    

if (pool_mlp_fc2_weight == nullptr) {
    throw std::runtime_error("Constant pool_mlp_fc2_weight was not set! Set the value with set_constant.");
}
    

if (pool_mlp_fc2_bias == nullptr) {
    throw std::runtime_error("Constant pool_mlp_fc2_bias was not set! Set the value with set_constant.");
}
    
     reshape_8_0 = split_7_0;
     reshape_9_0 = split_7_1;
     reshape_10_0 = split_7_2;
     reshape_23_0 = split_22_0;
     reshape_24_0 = split_22_1;
     reshape_25_0 = split_22_2;
     reshape_38_0 = split_37_0;
     reshape_39_0 = split_37_1;
     reshape_40_0 = split_37_2;
     reshape_53_0 = split_52_0;
     reshape_54_0 = split_52_1;
     reshape_55_0 = split_52_2;
     reshape_68_0 = split_67_0;
     reshape_69_0 = split_67_1;
     reshape_70_0 = split_67_2;
     reshape_83_0 = split_82_0;
     reshape_84_0 = split_82_1;
     reshape_85_0 = split_82_2;
     reshape_98_0 = split_97_0;
     reshape_99_0 = split_97_1;
     reshape_100_0 = split_97_2;
     reshape_113_0 = split_112_0;
     reshape_114_0 = split_112_1;
     reshape_115_0 = split_112_2;
     reshape_128_0 = split_127_0;
     reshape_129_0 = split_127_1;
     reshape_130_0 = split_127_2;
     reshape_143_0 = split_142_0;
     reshape_144_0 = split_142_1;
     reshape_145_0 = split_142_2;
     reshape_158_0 = split_157_0;
     reshape_159_0 = split_157_1;
     reshape_160_0 = split_157_2;
     reshape_173_0 = split_172_0;
     reshape_174_0 = split_172_1;
     reshape_175_0 = split_172_2;
     reshape_188_0 = split_187_0;
     reshape_189_0 = split_187_1;
     reshape_190_0 = split_187_2;
     reshape_203_0 = split_202_0;
     reshape_204_0 = split_202_1;
     reshape_205_0 = split_202_2;
     reshape_218_0 = split_217_0;
     reshape_219_0 = split_217_1;
     reshape_220_0 = split_217_2;
     reshape_233_0 = split_232_0;
     reshape_234_0 = split_232_1;
     reshape_235_0 = split_232_2;
     reshape_248_0 = split_247_0;
     reshape_249_0 = split_247_1;
     reshape_250_0 = split_247_2;
     reshape_263_0 = split_262_0;
     reshape_264_0 = split_262_1;
     reshape_265_0 = split_262_2;
     reshape_278_0 = split_277_0;
     reshape_279_0 = split_277_1;
     reshape_280_0 = split_277_2;
     reshape_293_0 = split_292_0;
     reshape_294_0 = split_292_1;
     reshape_295_0 = split_292_2;
     reshape_308_0 = split_307_0;
     reshape_309_0 = split_307_1;
     reshape_310_0 = split_307_2;
     reshape_323_0 = split_322_0;
     reshape_324_0 = split_322_1;
     reshape_325_0 = split_322_2;
     reshape_338_0 = split_337_0;
     reshape_339_0 = split_337_1;
     reshape_340_0 = split_337_2;
     reshape_353_0 = split_352_0;
     reshape_354_0 = split_352_1;
     reshape_355_0 = split_352_2;
     reshape_368_0 = split_367_0;
     reshape_369_0 = split_367_1;
     reshape_370_0 = split_367_2;
     reshape_383_0 = split_382_0;
     reshape_384_0 = split_382_1;
     reshape_385_0 = split_382_2;
     reshape_398_0 = split_397_0;
     reshape_399_0 = split_397_1;
     reshape_400_0 = split_397_2;
     squeeze_413_0 = split_412_0;
     squeeze_414_0 = split_412_1;
     reshape_426_0 = mem_eff_attention_418_0;
     output_0 = static_cast<decltype(output_0)>(params_[1].ptr);

if (output_0 == nullptr) {
    throw std::runtime_error("Constant output_0 was not set! Set the value with set_constant.");
}
    
    }

    void ResetConstants(uint8_t* constants) {
        /*
         * This can be called if we want to use a different piece of memory
         * for the constants to be consumed.
         */
            reshape_417_0 = reinterpret_cast<decltype(reshape_417_0)>(constants + 2658816);
    }

    void DeviceToDeviceCopies(StreamType stream) {
  
    }


    ///////////////////////////////////////////////////////////////////////////
    // default RunImpl implemenation
    void RunImpl(StreamType stream) {
        
  
  
    conv2d_bias_few_channels_0(
        input0,
        patch_embed_proj_weight,
        conv2d_bias_few_channels_0_0,

        patch_embed_proj_bias,

        global_workspace_,
        &input0_dim_0,
        &patch_embed_proj_weight_dim_0,
        &input0_dim_3,
        &patch_embed_proj_weight_dim_1,
        &patch_embed_proj_weight_dim_2,
        &input0_dim_1,
        &input0_dim_2,
        &input0_dim_0,
        &conv2d_bias_few_channels_0_0_dim_1,
        &conv2d_bias_few_channels_0_0_dim_2,
        14,
        1,
        0,
        14,
        1,
        0,
        stream
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
        int64_t fused_elementwise_425_n_elements = 8 * 729 * 1152;
        invoke_fused_elementwise_425(elementwise_2_0, conv2d_bias_few_channels_0_0,pos_emb_pos_emb,   fused_elementwise_425_n_elements, stream);
    }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_1_0_dim_0;

        M *= reshape_1_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_1_0_dim_2;

    
        layernorm_3(
           reshape_4_0, elementwise_2_0, encoder_layers_0_ln1_weight, encoder_layers_0_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_4_0,
        encoder_layers_0_mha_qkv_weight,

        encoder_layers_0_mha_qkv_bias,

        reshape_6_0,
        global_workspace_,
        1,

        &reshape_4_0_dim_0,

        &reshape_4_0_dim_1,


        &encoder_layers_0_mha_qkv_weight_dim_0,

        &encoder_layers_0_mha_qkv_weight_dim_1,


        &reshape_4_0_dim_0,

        &encoder_layers_0_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_7_0,
      split_7_1,
      split_7_2
    };


      int64_t *split_7_0_shape[] = {
        &split_7_0_dim_0, &reshape_6_0_dim_1, &reshape_6_0_dim_2, &reshape_6_0_dim_3, &reshape_6_0_dim_4
      };
      int64_t *split_7_1_shape[] = {
        &split_7_1_dim_0, &reshape_6_0_dim_1, &reshape_6_0_dim_2, &reshape_6_0_dim_3, &reshape_6_0_dim_4
      };
      int64_t *split_7_2_shape[] = {
        &split_7_2_dim_0, &reshape_6_0_dim_1, &reshape_6_0_dim_2, &reshape_6_0_dim_3, &reshape_6_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_7_0_shape, split_7_1_shape, split_7_2_shape
    };

    const int64_t reshape_6_0_shape[] = {
      reshape_6_0_dim_0, reshape_6_0_dim_1, reshape_6_0_dim_2, reshape_6_0_dim_3, reshape_6_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_6_0,
        reshape_6_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_7_0_dim_0,
        &reshape_6_0_dim_1,
        &reshape_6_0_dim_2,
        &reshape_6_0_dim_3,
        &reshape_6_0_dim_4,
        &reshape_8_0_dim_0,
        &reshape_8_0_dim_1,
        &reshape_8_0_dim_2,
        &reshape_8_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_7_1_dim_0,
        &reshape_6_0_dim_1,
        &reshape_6_0_dim_2,
        &reshape_6_0_dim_3,
        &reshape_6_0_dim_4,
        &reshape_9_0_dim_0,
        &reshape_9_0_dim_1,
        &reshape_9_0_dim_2,
        &reshape_9_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_7_2_dim_0,
        &reshape_6_0_dim_1,
        &reshape_6_0_dim_2,
        &reshape_6_0_dim_3,
        &reshape_6_0_dim_4,
        &reshape_10_0_dim_0,
        &reshape_10_0_dim_1,
        &reshape_10_0_dim_2,
        &reshape_10_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_11_0,
        reshape_8_0, reshape_9_0, reshape_10_0,
        &reshape_8_0_dim_0,
        &reshape_10_0_dim_2,
        &reshape_8_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_11_0,
        encoder_layers_0_mha_proj_weight,
        encoder_layers_0_mha_proj_bias,
        elementwise_2_0,

        reshape_14_0,
        global_workspace_,

     1,


        &reshape_12_0_dim_0,

        &reshape_12_0_dim_1,


        &encoder_layers_0_mha_proj_weight_dim_0,

        &encoder_layers_0_mha_proj_weight_dim_1,


        &reshape_12_0_dim_0,

        &encoder_layers_0_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_14_0_dim_0;

        M *= reshape_14_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_14_0_dim_2;

    
        layernorm_15(
           layernorm_15_0, reshape_14_0, encoder_layers_0_ln2_weight, encoder_layers_0_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_15_0,
        encoder_layers_0_mlp_fc1_weight,

        encoder_layers_0_mlp_fc1_bias,

        gemm_rcr_bias_gelu_16_0,
        global_workspace_,
        1,

        &reshape_14_0_dim_0,

        &reshape_14_0_dim_1,

        &reshape_14_0_dim_2,


        &encoder_layers_0_mlp_fc1_weight_dim_0,

        &encoder_layers_0_mlp_fc1_weight_dim_1,


        &reshape_14_0_dim_0,

        &reshape_14_0_dim_1,

        &encoder_layers_0_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_16_0,
        encoder_layers_0_mlp_fc2_weight,
        encoder_layers_0_mlp_fc2_bias,
        reshape_14_0,

        gemm_rcr_bias_add_17_0,
        global_workspace_,

     1,


        &reshape_14_0_dim_0,

        &reshape_14_0_dim_1,

        &encoder_layers_0_mlp_fc1_weight_dim_0,


        &encoder_layers_0_mlp_fc2_weight_dim_0,

        &encoder_layers_0_mlp_fc2_weight_dim_1,


        &reshape_14_0_dim_0,

        &reshape_14_0_dim_1,

        &encoder_layers_0_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_14_0_dim_0;

        M *= reshape_14_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_0_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_19_0, gemm_rcr_bias_add_17_0, encoder_layers_1_ln1_weight, encoder_layers_1_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_19_0,
        encoder_layers_1_mha_qkv_weight,

        encoder_layers_1_mha_qkv_bias,

        reshape_21_0,
        global_workspace_,
        1,

        &reshape_19_0_dim_0,

        &reshape_19_0_dim_1,


        &encoder_layers_1_mha_qkv_weight_dim_0,

        &encoder_layers_1_mha_qkv_weight_dim_1,


        &reshape_19_0_dim_0,

        &encoder_layers_1_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_22_0,
      split_22_2,
      split_22_1
    };


      int64_t *split_22_0_shape[] = {
        &split_22_0_dim_0, &reshape_21_0_dim_1, &reshape_21_0_dim_2, &reshape_21_0_dim_3, &reshape_21_0_dim_4
      };
      int64_t *split_22_2_shape[] = {
        &split_22_2_dim_0, &reshape_21_0_dim_1, &reshape_21_0_dim_2, &reshape_21_0_dim_3, &reshape_21_0_dim_4
      };
      int64_t *split_22_1_shape[] = {
        &split_22_1_dim_0, &reshape_21_0_dim_1, &reshape_21_0_dim_2, &reshape_21_0_dim_3, &reshape_21_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_22_0_shape, split_22_2_shape, split_22_1_shape
    };

    const int64_t reshape_21_0_shape[] = {
      reshape_21_0_dim_0, reshape_21_0_dim_1, reshape_21_0_dim_2, reshape_21_0_dim_3, reshape_21_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_21_0,
        reshape_21_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_22_0_dim_0,
        &reshape_21_0_dim_1,
        &reshape_21_0_dim_2,
        &reshape_21_0_dim_3,
        &reshape_21_0_dim_4,
        &reshape_23_0_dim_0,
        &reshape_23_0_dim_1,
        &reshape_23_0_dim_2,
        &reshape_23_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_22_1_dim_0,
        &reshape_21_0_dim_1,
        &reshape_21_0_dim_2,
        &reshape_21_0_dim_3,
        &reshape_21_0_dim_4,
        &reshape_24_0_dim_0,
        &reshape_24_0_dim_1,
        &reshape_24_0_dim_2,
        &reshape_24_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_22_2_dim_0,
        &reshape_21_0_dim_1,
        &reshape_21_0_dim_2,
        &reshape_21_0_dim_3,
        &reshape_21_0_dim_4,
        &reshape_25_0_dim_0,
        &reshape_25_0_dim_1,
        &reshape_25_0_dim_2,
        &reshape_25_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_26_0,
        reshape_23_0, reshape_25_0, reshape_24_0,
        &reshape_23_0_dim_0,
        &reshape_24_0_dim_2,
        &reshape_23_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_26_0,
        encoder_layers_1_mha_proj_weight,
        encoder_layers_1_mha_proj_bias,
        gemm_rcr_bias_add_17_0,

        reshape_29_0,
        global_workspace_,

     1,


        &reshape_27_0_dim_0,

        &reshape_27_0_dim_1,


        &encoder_layers_1_mha_proj_weight_dim_0,

        &encoder_layers_1_mha_proj_weight_dim_1,


        &reshape_27_0_dim_0,

        &encoder_layers_1_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_29_0_dim_0;

        M *= reshape_29_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_29_0_dim_2;

    
        layernorm_15(
           layernorm_30_0, reshape_29_0, encoder_layers_1_ln2_weight, encoder_layers_1_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_30_0,
        encoder_layers_1_mlp_fc1_weight,

        encoder_layers_1_mlp_fc1_bias,

        gemm_rcr_bias_gelu_31_0,
        global_workspace_,
        1,

        &reshape_29_0_dim_0,

        &reshape_29_0_dim_1,

        &reshape_29_0_dim_2,


        &encoder_layers_1_mlp_fc1_weight_dim_0,

        &encoder_layers_1_mlp_fc1_weight_dim_1,


        &reshape_29_0_dim_0,

        &reshape_29_0_dim_1,

        &encoder_layers_1_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_31_0,
        encoder_layers_1_mlp_fc2_weight,
        encoder_layers_1_mlp_fc2_bias,
        reshape_29_0,

        gemm_rcr_bias_add_32_0,
        global_workspace_,

     1,


        &reshape_29_0_dim_0,

        &reshape_29_0_dim_1,

        &encoder_layers_1_mlp_fc1_weight_dim_0,


        &encoder_layers_1_mlp_fc2_weight_dim_0,

        &encoder_layers_1_mlp_fc2_weight_dim_1,


        &reshape_29_0_dim_0,

        &reshape_29_0_dim_1,

        &encoder_layers_1_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_29_0_dim_0;

        M *= reshape_29_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_1_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_34_0, gemm_rcr_bias_add_32_0, encoder_layers_2_ln1_weight, encoder_layers_2_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_34_0,
        encoder_layers_2_mha_qkv_weight,

        encoder_layers_2_mha_qkv_bias,

        reshape_36_0,
        global_workspace_,
        1,

        &reshape_34_0_dim_0,

        &reshape_34_0_dim_1,


        &encoder_layers_2_mha_qkv_weight_dim_0,

        &encoder_layers_2_mha_qkv_weight_dim_1,


        &reshape_34_0_dim_0,

        &encoder_layers_2_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_37_1,
      split_37_0,
      split_37_2
    };


      int64_t *split_37_1_shape[] = {
        &split_37_1_dim_0, &reshape_36_0_dim_1, &reshape_36_0_dim_2, &reshape_36_0_dim_3, &reshape_36_0_dim_4
      };
      int64_t *split_37_0_shape[] = {
        &split_37_0_dim_0, &reshape_36_0_dim_1, &reshape_36_0_dim_2, &reshape_36_0_dim_3, &reshape_36_0_dim_4
      };
      int64_t *split_37_2_shape[] = {
        &split_37_2_dim_0, &reshape_36_0_dim_1, &reshape_36_0_dim_2, &reshape_36_0_dim_3, &reshape_36_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_37_1_shape, split_37_0_shape, split_37_2_shape
    };

    const int64_t reshape_36_0_shape[] = {
      reshape_36_0_dim_0, reshape_36_0_dim_1, reshape_36_0_dim_2, reshape_36_0_dim_3, reshape_36_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_36_0,
        reshape_36_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_37_0_dim_0,
        &reshape_36_0_dim_1,
        &reshape_36_0_dim_2,
        &reshape_36_0_dim_3,
        &reshape_36_0_dim_4,
        &reshape_38_0_dim_0,
        &reshape_38_0_dim_1,
        &reshape_38_0_dim_2,
        &reshape_38_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_37_1_dim_0,
        &reshape_36_0_dim_1,
        &reshape_36_0_dim_2,
        &reshape_36_0_dim_3,
        &reshape_36_0_dim_4,
        &reshape_39_0_dim_0,
        &reshape_39_0_dim_1,
        &reshape_39_0_dim_2,
        &reshape_39_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_37_2_dim_0,
        &reshape_36_0_dim_1,
        &reshape_36_0_dim_2,
        &reshape_36_0_dim_3,
        &reshape_36_0_dim_4,
        &reshape_40_0_dim_0,
        &reshape_40_0_dim_1,
        &reshape_40_0_dim_2,
        &reshape_40_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_41_0,
        reshape_39_0, reshape_38_0, reshape_40_0,
        &reshape_39_0_dim_0,
        &reshape_40_0_dim_2,
        &reshape_39_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_41_0,
        encoder_layers_2_mha_proj_weight,
        encoder_layers_2_mha_proj_bias,
        gemm_rcr_bias_add_32_0,

        reshape_44_0,
        global_workspace_,

     1,


        &reshape_42_0_dim_0,

        &reshape_42_0_dim_1,


        &encoder_layers_2_mha_proj_weight_dim_0,

        &encoder_layers_2_mha_proj_weight_dim_1,


        &reshape_42_0_dim_0,

        &encoder_layers_2_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_44_0_dim_0;

        M *= reshape_44_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_44_0_dim_2;

    
        layernorm_15(
           layernorm_45_0, reshape_44_0, encoder_layers_2_ln2_weight, encoder_layers_2_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_45_0,
        encoder_layers_2_mlp_fc1_weight,

        encoder_layers_2_mlp_fc1_bias,

        gemm_rcr_bias_gelu_46_0,
        global_workspace_,
        1,

        &reshape_44_0_dim_0,

        &reshape_44_0_dim_1,

        &reshape_44_0_dim_2,


        &encoder_layers_2_mlp_fc1_weight_dim_0,

        &encoder_layers_2_mlp_fc1_weight_dim_1,


        &reshape_44_0_dim_0,

        &reshape_44_0_dim_1,

        &encoder_layers_2_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_46_0,
        encoder_layers_2_mlp_fc2_weight,
        encoder_layers_2_mlp_fc2_bias,
        reshape_44_0,

        gemm_rcr_bias_add_47_0,
        global_workspace_,

     1,


        &reshape_44_0_dim_0,

        &reshape_44_0_dim_1,

        &encoder_layers_2_mlp_fc1_weight_dim_0,


        &encoder_layers_2_mlp_fc2_weight_dim_0,

        &encoder_layers_2_mlp_fc2_weight_dim_1,


        &reshape_44_0_dim_0,

        &reshape_44_0_dim_1,

        &encoder_layers_2_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_44_0_dim_0;

        M *= reshape_44_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_2_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_49_0, gemm_rcr_bias_add_47_0, encoder_layers_3_ln1_weight, encoder_layers_3_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_49_0,
        encoder_layers_3_mha_qkv_weight,

        encoder_layers_3_mha_qkv_bias,

        reshape_51_0,
        global_workspace_,
        1,

        &reshape_49_0_dim_0,

        &reshape_49_0_dim_1,


        &encoder_layers_3_mha_qkv_weight_dim_0,

        &encoder_layers_3_mha_qkv_weight_dim_1,


        &reshape_49_0_dim_0,

        &encoder_layers_3_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_52_2,
      split_52_1,
      split_52_0
    };


      int64_t *split_52_2_shape[] = {
        &split_52_2_dim_0, &reshape_51_0_dim_1, &reshape_51_0_dim_2, &reshape_51_0_dim_3, &reshape_51_0_dim_4
      };
      int64_t *split_52_1_shape[] = {
        &split_52_1_dim_0, &reshape_51_0_dim_1, &reshape_51_0_dim_2, &reshape_51_0_dim_3, &reshape_51_0_dim_4
      };
      int64_t *split_52_0_shape[] = {
        &split_52_0_dim_0, &reshape_51_0_dim_1, &reshape_51_0_dim_2, &reshape_51_0_dim_3, &reshape_51_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_52_2_shape, split_52_1_shape, split_52_0_shape
    };

    const int64_t reshape_51_0_shape[] = {
      reshape_51_0_dim_0, reshape_51_0_dim_1, reshape_51_0_dim_2, reshape_51_0_dim_3, reshape_51_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_51_0,
        reshape_51_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_52_0_dim_0,
        &reshape_51_0_dim_1,
        &reshape_51_0_dim_2,
        &reshape_51_0_dim_3,
        &reshape_51_0_dim_4,
        &reshape_53_0_dim_0,
        &reshape_53_0_dim_1,
        &reshape_53_0_dim_2,
        &reshape_53_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_52_1_dim_0,
        &reshape_51_0_dim_1,
        &reshape_51_0_dim_2,
        &reshape_51_0_dim_3,
        &reshape_51_0_dim_4,
        &reshape_54_0_dim_0,
        &reshape_54_0_dim_1,
        &reshape_54_0_dim_2,
        &reshape_54_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_52_2_dim_0,
        &reshape_51_0_dim_1,
        &reshape_51_0_dim_2,
        &reshape_51_0_dim_3,
        &reshape_51_0_dim_4,
        &reshape_55_0_dim_0,
        &reshape_55_0_dim_1,
        &reshape_55_0_dim_2,
        &reshape_55_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_56_0,
        reshape_55_0, reshape_54_0, reshape_53_0,
        &reshape_55_0_dim_0,
        &reshape_53_0_dim_2,
        &reshape_55_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_56_0,
        encoder_layers_3_mha_proj_weight,
        encoder_layers_3_mha_proj_bias,
        gemm_rcr_bias_add_47_0,

        reshape_59_0,
        global_workspace_,

     1,


        &reshape_57_0_dim_0,

        &reshape_57_0_dim_1,


        &encoder_layers_3_mha_proj_weight_dim_0,

        &encoder_layers_3_mha_proj_weight_dim_1,


        &reshape_57_0_dim_0,

        &encoder_layers_3_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_59_0_dim_0;

        M *= reshape_59_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_59_0_dim_2;

    
        layernorm_15(
           layernorm_60_0, reshape_59_0, encoder_layers_3_ln2_weight, encoder_layers_3_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_60_0,
        encoder_layers_3_mlp_fc1_weight,

        encoder_layers_3_mlp_fc1_bias,

        gemm_rcr_bias_gelu_61_0,
        global_workspace_,
        1,

        &reshape_59_0_dim_0,

        &reshape_59_0_dim_1,

        &reshape_59_0_dim_2,


        &encoder_layers_3_mlp_fc1_weight_dim_0,

        &encoder_layers_3_mlp_fc1_weight_dim_1,


        &reshape_59_0_dim_0,

        &reshape_59_0_dim_1,

        &encoder_layers_3_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_61_0,
        encoder_layers_3_mlp_fc2_weight,
        encoder_layers_3_mlp_fc2_bias,
        reshape_59_0,

        gemm_rcr_bias_add_62_0,
        global_workspace_,

     1,


        &reshape_59_0_dim_0,

        &reshape_59_0_dim_1,

        &encoder_layers_3_mlp_fc1_weight_dim_0,


        &encoder_layers_3_mlp_fc2_weight_dim_0,

        &encoder_layers_3_mlp_fc2_weight_dim_1,


        &reshape_59_0_dim_0,

        &reshape_59_0_dim_1,

        &encoder_layers_3_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_59_0_dim_0;

        M *= reshape_59_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_3_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_64_0, gemm_rcr_bias_add_62_0, encoder_layers_4_ln1_weight, encoder_layers_4_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_64_0,
        encoder_layers_4_mha_qkv_weight,

        encoder_layers_4_mha_qkv_bias,

        reshape_66_0,
        global_workspace_,
        1,

        &reshape_64_0_dim_0,

        &reshape_64_0_dim_1,


        &encoder_layers_4_mha_qkv_weight_dim_0,

        &encoder_layers_4_mha_qkv_weight_dim_1,


        &reshape_64_0_dim_0,

        &encoder_layers_4_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_67_0,
      split_67_1,
      split_67_2
    };


      int64_t *split_67_0_shape[] = {
        &split_67_0_dim_0, &reshape_66_0_dim_1, &reshape_66_0_dim_2, &reshape_66_0_dim_3, &reshape_66_0_dim_4
      };
      int64_t *split_67_1_shape[] = {
        &split_67_1_dim_0, &reshape_66_0_dim_1, &reshape_66_0_dim_2, &reshape_66_0_dim_3, &reshape_66_0_dim_4
      };
      int64_t *split_67_2_shape[] = {
        &split_67_2_dim_0, &reshape_66_0_dim_1, &reshape_66_0_dim_2, &reshape_66_0_dim_3, &reshape_66_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_67_0_shape, split_67_1_shape, split_67_2_shape
    };

    const int64_t reshape_66_0_shape[] = {
      reshape_66_0_dim_0, reshape_66_0_dim_1, reshape_66_0_dim_2, reshape_66_0_dim_3, reshape_66_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_66_0,
        reshape_66_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_67_0_dim_0,
        &reshape_66_0_dim_1,
        &reshape_66_0_dim_2,
        &reshape_66_0_dim_3,
        &reshape_66_0_dim_4,
        &reshape_68_0_dim_0,
        &reshape_68_0_dim_1,
        &reshape_68_0_dim_2,
        &reshape_68_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_67_1_dim_0,
        &reshape_66_0_dim_1,
        &reshape_66_0_dim_2,
        &reshape_66_0_dim_3,
        &reshape_66_0_dim_4,
        &reshape_69_0_dim_0,
        &reshape_69_0_dim_1,
        &reshape_69_0_dim_2,
        &reshape_69_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_67_2_dim_0,
        &reshape_66_0_dim_1,
        &reshape_66_0_dim_2,
        &reshape_66_0_dim_3,
        &reshape_66_0_dim_4,
        &reshape_70_0_dim_0,
        &reshape_70_0_dim_1,
        &reshape_70_0_dim_2,
        &reshape_70_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_71_0,
        reshape_68_0, reshape_69_0, reshape_70_0,
        &reshape_68_0_dim_0,
        &reshape_70_0_dim_2,
        &reshape_68_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_71_0,
        encoder_layers_4_mha_proj_weight,
        encoder_layers_4_mha_proj_bias,
        gemm_rcr_bias_add_62_0,

        reshape_74_0,
        global_workspace_,

     1,


        &reshape_72_0_dim_0,

        &reshape_72_0_dim_1,


        &encoder_layers_4_mha_proj_weight_dim_0,

        &encoder_layers_4_mha_proj_weight_dim_1,


        &reshape_72_0_dim_0,

        &encoder_layers_4_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_74_0_dim_0;

        M *= reshape_74_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_74_0_dim_2;

    
        layernorm_15(
           layernorm_75_0, reshape_74_0, encoder_layers_4_ln2_weight, encoder_layers_4_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_75_0,
        encoder_layers_4_mlp_fc1_weight,

        encoder_layers_4_mlp_fc1_bias,

        gemm_rcr_bias_gelu_76_0,
        global_workspace_,
        1,

        &reshape_74_0_dim_0,

        &reshape_74_0_dim_1,

        &reshape_74_0_dim_2,


        &encoder_layers_4_mlp_fc1_weight_dim_0,

        &encoder_layers_4_mlp_fc1_weight_dim_1,


        &reshape_74_0_dim_0,

        &reshape_74_0_dim_1,

        &encoder_layers_4_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_76_0,
        encoder_layers_4_mlp_fc2_weight,
        encoder_layers_4_mlp_fc2_bias,
        reshape_74_0,

        gemm_rcr_bias_add_77_0,
        global_workspace_,

     1,


        &reshape_74_0_dim_0,

        &reshape_74_0_dim_1,

        &encoder_layers_4_mlp_fc1_weight_dim_0,


        &encoder_layers_4_mlp_fc2_weight_dim_0,

        &encoder_layers_4_mlp_fc2_weight_dim_1,


        &reshape_74_0_dim_0,

        &reshape_74_0_dim_1,

        &encoder_layers_4_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_74_0_dim_0;

        M *= reshape_74_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_4_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_79_0, gemm_rcr_bias_add_77_0, encoder_layers_5_ln1_weight, encoder_layers_5_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_79_0,
        encoder_layers_5_mha_qkv_weight,

        encoder_layers_5_mha_qkv_bias,

        reshape_81_0,
        global_workspace_,
        1,

        &reshape_79_0_dim_0,

        &reshape_79_0_dim_1,


        &encoder_layers_5_mha_qkv_weight_dim_0,

        &encoder_layers_5_mha_qkv_weight_dim_1,


        &reshape_79_0_dim_0,

        &encoder_layers_5_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_82_0,
      split_82_1,
      split_82_2
    };


      int64_t *split_82_0_shape[] = {
        &split_82_0_dim_0, &reshape_81_0_dim_1, &reshape_81_0_dim_2, &reshape_81_0_dim_3, &reshape_81_0_dim_4
      };
      int64_t *split_82_1_shape[] = {
        &split_82_1_dim_0, &reshape_81_0_dim_1, &reshape_81_0_dim_2, &reshape_81_0_dim_3, &reshape_81_0_dim_4
      };
      int64_t *split_82_2_shape[] = {
        &split_82_2_dim_0, &reshape_81_0_dim_1, &reshape_81_0_dim_2, &reshape_81_0_dim_3, &reshape_81_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_82_0_shape, split_82_1_shape, split_82_2_shape
    };

    const int64_t reshape_81_0_shape[] = {
      reshape_81_0_dim_0, reshape_81_0_dim_1, reshape_81_0_dim_2, reshape_81_0_dim_3, reshape_81_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_81_0,
        reshape_81_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_82_0_dim_0,
        &reshape_81_0_dim_1,
        &reshape_81_0_dim_2,
        &reshape_81_0_dim_3,
        &reshape_81_0_dim_4,
        &reshape_83_0_dim_0,
        &reshape_83_0_dim_1,
        &reshape_83_0_dim_2,
        &reshape_83_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_82_1_dim_0,
        &reshape_81_0_dim_1,
        &reshape_81_0_dim_2,
        &reshape_81_0_dim_3,
        &reshape_81_0_dim_4,
        &reshape_84_0_dim_0,
        &reshape_84_0_dim_1,
        &reshape_84_0_dim_2,
        &reshape_84_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_82_2_dim_0,
        &reshape_81_0_dim_1,
        &reshape_81_0_dim_2,
        &reshape_81_0_dim_3,
        &reshape_81_0_dim_4,
        &reshape_85_0_dim_0,
        &reshape_85_0_dim_1,
        &reshape_85_0_dim_2,
        &reshape_85_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_86_0,
        reshape_83_0, reshape_84_0, reshape_85_0,
        &reshape_83_0_dim_0,
        &reshape_85_0_dim_2,
        &reshape_83_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_86_0,
        encoder_layers_5_mha_proj_weight,
        encoder_layers_5_mha_proj_bias,
        gemm_rcr_bias_add_77_0,

        reshape_89_0,
        global_workspace_,

     1,


        &reshape_87_0_dim_0,

        &reshape_87_0_dim_1,


        &encoder_layers_5_mha_proj_weight_dim_0,

        &encoder_layers_5_mha_proj_weight_dim_1,


        &reshape_87_0_dim_0,

        &encoder_layers_5_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_89_0_dim_0;

        M *= reshape_89_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_89_0_dim_2;

    
        layernorm_15(
           layernorm_90_0, reshape_89_0, encoder_layers_5_ln2_weight, encoder_layers_5_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_90_0,
        encoder_layers_5_mlp_fc1_weight,

        encoder_layers_5_mlp_fc1_bias,

        gemm_rcr_bias_gelu_91_0,
        global_workspace_,
        1,

        &reshape_89_0_dim_0,

        &reshape_89_0_dim_1,

        &reshape_89_0_dim_2,


        &encoder_layers_5_mlp_fc1_weight_dim_0,

        &encoder_layers_5_mlp_fc1_weight_dim_1,


        &reshape_89_0_dim_0,

        &reshape_89_0_dim_1,

        &encoder_layers_5_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_91_0,
        encoder_layers_5_mlp_fc2_weight,
        encoder_layers_5_mlp_fc2_bias,
        reshape_89_0,

        gemm_rcr_bias_add_92_0,
        global_workspace_,

     1,


        &reshape_89_0_dim_0,

        &reshape_89_0_dim_1,

        &encoder_layers_5_mlp_fc1_weight_dim_0,


        &encoder_layers_5_mlp_fc2_weight_dim_0,

        &encoder_layers_5_mlp_fc2_weight_dim_1,


        &reshape_89_0_dim_0,

        &reshape_89_0_dim_1,

        &encoder_layers_5_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_89_0_dim_0;

        M *= reshape_89_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_5_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_94_0, gemm_rcr_bias_add_92_0, encoder_layers_6_ln1_weight, encoder_layers_6_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_94_0,
        encoder_layers_6_mha_qkv_weight,

        encoder_layers_6_mha_qkv_bias,

        reshape_96_0,
        global_workspace_,
        1,

        &reshape_94_0_dim_0,

        &reshape_94_0_dim_1,


        &encoder_layers_6_mha_qkv_weight_dim_0,

        &encoder_layers_6_mha_qkv_weight_dim_1,


        &reshape_94_0_dim_0,

        &encoder_layers_6_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_97_2,
      split_97_0,
      split_97_1
    };


      int64_t *split_97_2_shape[] = {
        &split_97_2_dim_0, &reshape_96_0_dim_1, &reshape_96_0_dim_2, &reshape_96_0_dim_3, &reshape_96_0_dim_4
      };
      int64_t *split_97_0_shape[] = {
        &split_97_0_dim_0, &reshape_96_0_dim_1, &reshape_96_0_dim_2, &reshape_96_0_dim_3, &reshape_96_0_dim_4
      };
      int64_t *split_97_1_shape[] = {
        &split_97_1_dim_0, &reshape_96_0_dim_1, &reshape_96_0_dim_2, &reshape_96_0_dim_3, &reshape_96_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_97_2_shape, split_97_0_shape, split_97_1_shape
    };

    const int64_t reshape_96_0_shape[] = {
      reshape_96_0_dim_0, reshape_96_0_dim_1, reshape_96_0_dim_2, reshape_96_0_dim_3, reshape_96_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_96_0,
        reshape_96_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_97_0_dim_0,
        &reshape_96_0_dim_1,
        &reshape_96_0_dim_2,
        &reshape_96_0_dim_3,
        &reshape_96_0_dim_4,
        &reshape_98_0_dim_0,
        &reshape_98_0_dim_1,
        &reshape_98_0_dim_2,
        &reshape_98_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_97_1_dim_0,
        &reshape_96_0_dim_1,
        &reshape_96_0_dim_2,
        &reshape_96_0_dim_3,
        &reshape_96_0_dim_4,
        &reshape_99_0_dim_0,
        &reshape_99_0_dim_1,
        &reshape_99_0_dim_2,
        &reshape_99_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_97_2_dim_0,
        &reshape_96_0_dim_1,
        &reshape_96_0_dim_2,
        &reshape_96_0_dim_3,
        &reshape_96_0_dim_4,
        &reshape_100_0_dim_0,
        &reshape_100_0_dim_1,
        &reshape_100_0_dim_2,
        &reshape_100_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_101_0,
        reshape_100_0, reshape_98_0, reshape_99_0,
        &reshape_100_0_dim_0,
        &reshape_99_0_dim_2,
        &reshape_100_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_101_0,
        encoder_layers_6_mha_proj_weight,
        encoder_layers_6_mha_proj_bias,
        gemm_rcr_bias_add_92_0,

        reshape_104_0,
        global_workspace_,

     1,


        &reshape_102_0_dim_0,

        &reshape_102_0_dim_1,


        &encoder_layers_6_mha_proj_weight_dim_0,

        &encoder_layers_6_mha_proj_weight_dim_1,


        &reshape_102_0_dim_0,

        &encoder_layers_6_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_104_0_dim_0;

        M *= reshape_104_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_104_0_dim_2;

    
        layernorm_15(
           layernorm_105_0, reshape_104_0, encoder_layers_6_ln2_weight, encoder_layers_6_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_105_0,
        encoder_layers_6_mlp_fc1_weight,

        encoder_layers_6_mlp_fc1_bias,

        gemm_rcr_bias_gelu_106_0,
        global_workspace_,
        1,

        &reshape_104_0_dim_0,

        &reshape_104_0_dim_1,

        &reshape_104_0_dim_2,


        &encoder_layers_6_mlp_fc1_weight_dim_0,

        &encoder_layers_6_mlp_fc1_weight_dim_1,


        &reshape_104_0_dim_0,

        &reshape_104_0_dim_1,

        &encoder_layers_6_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_106_0,
        encoder_layers_6_mlp_fc2_weight,
        encoder_layers_6_mlp_fc2_bias,
        reshape_104_0,

        gemm_rcr_bias_add_107_0,
        global_workspace_,

     1,


        &reshape_104_0_dim_0,

        &reshape_104_0_dim_1,

        &encoder_layers_6_mlp_fc1_weight_dim_0,


        &encoder_layers_6_mlp_fc2_weight_dim_0,

        &encoder_layers_6_mlp_fc2_weight_dim_1,


        &reshape_104_0_dim_0,

        &reshape_104_0_dim_1,

        &encoder_layers_6_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_104_0_dim_0;

        M *= reshape_104_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_6_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_109_0, gemm_rcr_bias_add_107_0, encoder_layers_7_ln1_weight, encoder_layers_7_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_109_0,
        encoder_layers_7_mha_qkv_weight,

        encoder_layers_7_mha_qkv_bias,

        reshape_111_0,
        global_workspace_,
        1,

        &reshape_109_0_dim_0,

        &reshape_109_0_dim_1,


        &encoder_layers_7_mha_qkv_weight_dim_0,

        &encoder_layers_7_mha_qkv_weight_dim_1,


        &reshape_109_0_dim_0,

        &encoder_layers_7_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_112_0,
      split_112_1,
      split_112_2
    };


      int64_t *split_112_0_shape[] = {
        &split_112_0_dim_0, &reshape_111_0_dim_1, &reshape_111_0_dim_2, &reshape_111_0_dim_3, &reshape_111_0_dim_4
      };
      int64_t *split_112_1_shape[] = {
        &split_112_1_dim_0, &reshape_111_0_dim_1, &reshape_111_0_dim_2, &reshape_111_0_dim_3, &reshape_111_0_dim_4
      };
      int64_t *split_112_2_shape[] = {
        &split_112_2_dim_0, &reshape_111_0_dim_1, &reshape_111_0_dim_2, &reshape_111_0_dim_3, &reshape_111_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_112_0_shape, split_112_1_shape, split_112_2_shape
    };

    const int64_t reshape_111_0_shape[] = {
      reshape_111_0_dim_0, reshape_111_0_dim_1, reshape_111_0_dim_2, reshape_111_0_dim_3, reshape_111_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_111_0,
        reshape_111_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_112_0_dim_0,
        &reshape_111_0_dim_1,
        &reshape_111_0_dim_2,
        &reshape_111_0_dim_3,
        &reshape_111_0_dim_4,
        &reshape_113_0_dim_0,
        &reshape_113_0_dim_1,
        &reshape_113_0_dim_2,
        &reshape_113_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_112_1_dim_0,
        &reshape_111_0_dim_1,
        &reshape_111_0_dim_2,
        &reshape_111_0_dim_3,
        &reshape_111_0_dim_4,
        &reshape_114_0_dim_0,
        &reshape_114_0_dim_1,
        &reshape_114_0_dim_2,
        &reshape_114_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_112_2_dim_0,
        &reshape_111_0_dim_1,
        &reshape_111_0_dim_2,
        &reshape_111_0_dim_3,
        &reshape_111_0_dim_4,
        &reshape_115_0_dim_0,
        &reshape_115_0_dim_1,
        &reshape_115_0_dim_2,
        &reshape_115_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_116_0,
        reshape_113_0, reshape_114_0, reshape_115_0,
        &reshape_113_0_dim_0,
        &reshape_115_0_dim_2,
        &reshape_113_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_116_0,
        encoder_layers_7_mha_proj_weight,
        encoder_layers_7_mha_proj_bias,
        gemm_rcr_bias_add_107_0,

        reshape_119_0,
        global_workspace_,

     1,


        &reshape_117_0_dim_0,

        &reshape_117_0_dim_1,


        &encoder_layers_7_mha_proj_weight_dim_0,

        &encoder_layers_7_mha_proj_weight_dim_1,


        &reshape_117_0_dim_0,

        &encoder_layers_7_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_119_0_dim_0;

        M *= reshape_119_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_119_0_dim_2;

    
        layernorm_15(
           layernorm_120_0, reshape_119_0, encoder_layers_7_ln2_weight, encoder_layers_7_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_120_0,
        encoder_layers_7_mlp_fc1_weight,

        encoder_layers_7_mlp_fc1_bias,

        gemm_rcr_bias_gelu_121_0,
        global_workspace_,
        1,

        &reshape_119_0_dim_0,

        &reshape_119_0_dim_1,

        &reshape_119_0_dim_2,


        &encoder_layers_7_mlp_fc1_weight_dim_0,

        &encoder_layers_7_mlp_fc1_weight_dim_1,


        &reshape_119_0_dim_0,

        &reshape_119_0_dim_1,

        &encoder_layers_7_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_121_0,
        encoder_layers_7_mlp_fc2_weight,
        encoder_layers_7_mlp_fc2_bias,
        reshape_119_0,

        gemm_rcr_bias_add_122_0,
        global_workspace_,

     1,


        &reshape_119_0_dim_0,

        &reshape_119_0_dim_1,

        &encoder_layers_7_mlp_fc1_weight_dim_0,


        &encoder_layers_7_mlp_fc2_weight_dim_0,

        &encoder_layers_7_mlp_fc2_weight_dim_1,


        &reshape_119_0_dim_0,

        &reshape_119_0_dim_1,

        &encoder_layers_7_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_119_0_dim_0;

        M *= reshape_119_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_7_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_124_0, gemm_rcr_bias_add_122_0, encoder_layers_8_ln1_weight, encoder_layers_8_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_124_0,
        encoder_layers_8_mha_qkv_weight,

        encoder_layers_8_mha_qkv_bias,

        reshape_126_0,
        global_workspace_,
        1,

        &reshape_124_0_dim_0,

        &reshape_124_0_dim_1,


        &encoder_layers_8_mha_qkv_weight_dim_0,

        &encoder_layers_8_mha_qkv_weight_dim_1,


        &reshape_124_0_dim_0,

        &encoder_layers_8_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_127_1,
      split_127_0,
      split_127_2
    };


      int64_t *split_127_1_shape[] = {
        &split_127_1_dim_0, &reshape_126_0_dim_1, &reshape_126_0_dim_2, &reshape_126_0_dim_3, &reshape_126_0_dim_4
      };
      int64_t *split_127_0_shape[] = {
        &split_127_0_dim_0, &reshape_126_0_dim_1, &reshape_126_0_dim_2, &reshape_126_0_dim_3, &reshape_126_0_dim_4
      };
      int64_t *split_127_2_shape[] = {
        &split_127_2_dim_0, &reshape_126_0_dim_1, &reshape_126_0_dim_2, &reshape_126_0_dim_3, &reshape_126_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_127_1_shape, split_127_0_shape, split_127_2_shape
    };

    const int64_t reshape_126_0_shape[] = {
      reshape_126_0_dim_0, reshape_126_0_dim_1, reshape_126_0_dim_2, reshape_126_0_dim_3, reshape_126_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_126_0,
        reshape_126_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_127_0_dim_0,
        &reshape_126_0_dim_1,
        &reshape_126_0_dim_2,
        &reshape_126_0_dim_3,
        &reshape_126_0_dim_4,
        &reshape_128_0_dim_0,
        &reshape_128_0_dim_1,
        &reshape_128_0_dim_2,
        &reshape_128_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_127_1_dim_0,
        &reshape_126_0_dim_1,
        &reshape_126_0_dim_2,
        &reshape_126_0_dim_3,
        &reshape_126_0_dim_4,
        &reshape_129_0_dim_0,
        &reshape_129_0_dim_1,
        &reshape_129_0_dim_2,
        &reshape_129_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_127_2_dim_0,
        &reshape_126_0_dim_1,
        &reshape_126_0_dim_2,
        &reshape_126_0_dim_3,
        &reshape_126_0_dim_4,
        &reshape_130_0_dim_0,
        &reshape_130_0_dim_1,
        &reshape_130_0_dim_2,
        &reshape_130_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_131_0,
        reshape_129_0, reshape_128_0, reshape_130_0,
        &reshape_129_0_dim_0,
        &reshape_130_0_dim_2,
        &reshape_129_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_131_0,
        encoder_layers_8_mha_proj_weight,
        encoder_layers_8_mha_proj_bias,
        gemm_rcr_bias_add_122_0,

        reshape_134_0,
        global_workspace_,

     1,


        &reshape_132_0_dim_0,

        &reshape_132_0_dim_1,


        &encoder_layers_8_mha_proj_weight_dim_0,

        &encoder_layers_8_mha_proj_weight_dim_1,


        &reshape_132_0_dim_0,

        &encoder_layers_8_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_134_0_dim_0;

        M *= reshape_134_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_134_0_dim_2;

    
        layernorm_15(
           layernorm_135_0, reshape_134_0, encoder_layers_8_ln2_weight, encoder_layers_8_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_135_0,
        encoder_layers_8_mlp_fc1_weight,

        encoder_layers_8_mlp_fc1_bias,

        gemm_rcr_bias_gelu_136_0,
        global_workspace_,
        1,

        &reshape_134_0_dim_0,

        &reshape_134_0_dim_1,

        &reshape_134_0_dim_2,


        &encoder_layers_8_mlp_fc1_weight_dim_0,

        &encoder_layers_8_mlp_fc1_weight_dim_1,


        &reshape_134_0_dim_0,

        &reshape_134_0_dim_1,

        &encoder_layers_8_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_136_0,
        encoder_layers_8_mlp_fc2_weight,
        encoder_layers_8_mlp_fc2_bias,
        reshape_134_0,

        gemm_rcr_bias_add_137_0,
        global_workspace_,

     1,


        &reshape_134_0_dim_0,

        &reshape_134_0_dim_1,

        &encoder_layers_8_mlp_fc1_weight_dim_0,


        &encoder_layers_8_mlp_fc2_weight_dim_0,

        &encoder_layers_8_mlp_fc2_weight_dim_1,


        &reshape_134_0_dim_0,

        &reshape_134_0_dim_1,

        &encoder_layers_8_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_134_0_dim_0;

        M *= reshape_134_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_8_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_139_0, gemm_rcr_bias_add_137_0, encoder_layers_9_ln1_weight, encoder_layers_9_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_139_0,
        encoder_layers_9_mha_qkv_weight,

        encoder_layers_9_mha_qkv_bias,

        reshape_141_0,
        global_workspace_,
        1,

        &reshape_139_0_dim_0,

        &reshape_139_0_dim_1,


        &encoder_layers_9_mha_qkv_weight_dim_0,

        &encoder_layers_9_mha_qkv_weight_dim_1,


        &reshape_139_0_dim_0,

        &encoder_layers_9_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_142_2,
      split_142_1,
      split_142_0
    };


      int64_t *split_142_2_shape[] = {
        &split_142_2_dim_0, &reshape_141_0_dim_1, &reshape_141_0_dim_2, &reshape_141_0_dim_3, &reshape_141_0_dim_4
      };
      int64_t *split_142_1_shape[] = {
        &split_142_1_dim_0, &reshape_141_0_dim_1, &reshape_141_0_dim_2, &reshape_141_0_dim_3, &reshape_141_0_dim_4
      };
      int64_t *split_142_0_shape[] = {
        &split_142_0_dim_0, &reshape_141_0_dim_1, &reshape_141_0_dim_2, &reshape_141_0_dim_3, &reshape_141_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_142_2_shape, split_142_1_shape, split_142_0_shape
    };

    const int64_t reshape_141_0_shape[] = {
      reshape_141_0_dim_0, reshape_141_0_dim_1, reshape_141_0_dim_2, reshape_141_0_dim_3, reshape_141_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_141_0,
        reshape_141_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_142_0_dim_0,
        &reshape_141_0_dim_1,
        &reshape_141_0_dim_2,
        &reshape_141_0_dim_3,
        &reshape_141_0_dim_4,
        &reshape_143_0_dim_0,
        &reshape_143_0_dim_1,
        &reshape_143_0_dim_2,
        &reshape_143_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_142_1_dim_0,
        &reshape_141_0_dim_1,
        &reshape_141_0_dim_2,
        &reshape_141_0_dim_3,
        &reshape_141_0_dim_4,
        &reshape_144_0_dim_0,
        &reshape_144_0_dim_1,
        &reshape_144_0_dim_2,
        &reshape_144_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_142_2_dim_0,
        &reshape_141_0_dim_1,
        &reshape_141_0_dim_2,
        &reshape_141_0_dim_3,
        &reshape_141_0_dim_4,
        &reshape_145_0_dim_0,
        &reshape_145_0_dim_1,
        &reshape_145_0_dim_2,
        &reshape_145_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_146_0,
        reshape_145_0, reshape_144_0, reshape_143_0,
        &reshape_145_0_dim_0,
        &reshape_143_0_dim_2,
        &reshape_145_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_146_0,
        encoder_layers_9_mha_proj_weight,
        encoder_layers_9_mha_proj_bias,
        gemm_rcr_bias_add_137_0,

        reshape_149_0,
        global_workspace_,

     1,


        &reshape_147_0_dim_0,

        &reshape_147_0_dim_1,


        &encoder_layers_9_mha_proj_weight_dim_0,

        &encoder_layers_9_mha_proj_weight_dim_1,


        &reshape_147_0_dim_0,

        &encoder_layers_9_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_149_0_dim_0;

        M *= reshape_149_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_149_0_dim_2;

    
        layernorm_15(
           layernorm_150_0, reshape_149_0, encoder_layers_9_ln2_weight, encoder_layers_9_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_150_0,
        encoder_layers_9_mlp_fc1_weight,

        encoder_layers_9_mlp_fc1_bias,

        gemm_rcr_bias_gelu_151_0,
        global_workspace_,
        1,

        &reshape_149_0_dim_0,

        &reshape_149_0_dim_1,

        &reshape_149_0_dim_2,


        &encoder_layers_9_mlp_fc1_weight_dim_0,

        &encoder_layers_9_mlp_fc1_weight_dim_1,


        &reshape_149_0_dim_0,

        &reshape_149_0_dim_1,

        &encoder_layers_9_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_151_0,
        encoder_layers_9_mlp_fc2_weight,
        encoder_layers_9_mlp_fc2_bias,
        reshape_149_0,

        gemm_rcr_bias_add_152_0,
        global_workspace_,

     1,


        &reshape_149_0_dim_0,

        &reshape_149_0_dim_1,

        &encoder_layers_9_mlp_fc1_weight_dim_0,


        &encoder_layers_9_mlp_fc2_weight_dim_0,

        &encoder_layers_9_mlp_fc2_weight_dim_1,


        &reshape_149_0_dim_0,

        &reshape_149_0_dim_1,

        &encoder_layers_9_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_149_0_dim_0;

        M *= reshape_149_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_9_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_154_0, gemm_rcr_bias_add_152_0, encoder_layers_10_ln1_weight, encoder_layers_10_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_154_0,
        encoder_layers_10_mha_qkv_weight,

        encoder_layers_10_mha_qkv_bias,

        reshape_156_0,
        global_workspace_,
        1,

        &reshape_154_0_dim_0,

        &reshape_154_0_dim_1,


        &encoder_layers_10_mha_qkv_weight_dim_0,

        &encoder_layers_10_mha_qkv_weight_dim_1,


        &reshape_154_0_dim_0,

        &encoder_layers_10_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_157_1,
      split_157_2,
      split_157_0
    };


      int64_t *split_157_1_shape[] = {
        &split_157_1_dim_0, &reshape_156_0_dim_1, &reshape_156_0_dim_2, &reshape_156_0_dim_3, &reshape_156_0_dim_4
      };
      int64_t *split_157_2_shape[] = {
        &split_157_2_dim_0, &reshape_156_0_dim_1, &reshape_156_0_dim_2, &reshape_156_0_dim_3, &reshape_156_0_dim_4
      };
      int64_t *split_157_0_shape[] = {
        &split_157_0_dim_0, &reshape_156_0_dim_1, &reshape_156_0_dim_2, &reshape_156_0_dim_3, &reshape_156_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_157_1_shape, split_157_2_shape, split_157_0_shape
    };

    const int64_t reshape_156_0_shape[] = {
      reshape_156_0_dim_0, reshape_156_0_dim_1, reshape_156_0_dim_2, reshape_156_0_dim_3, reshape_156_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_156_0,
        reshape_156_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_157_0_dim_0,
        &reshape_156_0_dim_1,
        &reshape_156_0_dim_2,
        &reshape_156_0_dim_3,
        &reshape_156_0_dim_4,
        &reshape_158_0_dim_0,
        &reshape_158_0_dim_1,
        &reshape_158_0_dim_2,
        &reshape_158_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_157_1_dim_0,
        &reshape_156_0_dim_1,
        &reshape_156_0_dim_2,
        &reshape_156_0_dim_3,
        &reshape_156_0_dim_4,
        &reshape_159_0_dim_0,
        &reshape_159_0_dim_1,
        &reshape_159_0_dim_2,
        &reshape_159_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_157_2_dim_0,
        &reshape_156_0_dim_1,
        &reshape_156_0_dim_2,
        &reshape_156_0_dim_3,
        &reshape_156_0_dim_4,
        &reshape_160_0_dim_0,
        &reshape_160_0_dim_1,
        &reshape_160_0_dim_2,
        &reshape_160_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_161_0,
        reshape_159_0, reshape_160_0, reshape_158_0,
        &reshape_159_0_dim_0,
        &reshape_158_0_dim_2,
        &reshape_159_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_161_0,
        encoder_layers_10_mha_proj_weight,
        encoder_layers_10_mha_proj_bias,
        gemm_rcr_bias_add_152_0,

        reshape_164_0,
        global_workspace_,

     1,


        &reshape_162_0_dim_0,

        &reshape_162_0_dim_1,


        &encoder_layers_10_mha_proj_weight_dim_0,

        &encoder_layers_10_mha_proj_weight_dim_1,


        &reshape_162_0_dim_0,

        &encoder_layers_10_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_164_0_dim_0;

        M *= reshape_164_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_164_0_dim_2;

    
        layernorm_15(
           layernorm_165_0, reshape_164_0, encoder_layers_10_ln2_weight, encoder_layers_10_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_165_0,
        encoder_layers_10_mlp_fc1_weight,

        encoder_layers_10_mlp_fc1_bias,

        gemm_rcr_bias_gelu_166_0,
        global_workspace_,
        1,

        &reshape_164_0_dim_0,

        &reshape_164_0_dim_1,

        &reshape_164_0_dim_2,


        &encoder_layers_10_mlp_fc1_weight_dim_0,

        &encoder_layers_10_mlp_fc1_weight_dim_1,


        &reshape_164_0_dim_0,

        &reshape_164_0_dim_1,

        &encoder_layers_10_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_166_0,
        encoder_layers_10_mlp_fc2_weight,
        encoder_layers_10_mlp_fc2_bias,
        reshape_164_0,

        gemm_rcr_bias_add_167_0,
        global_workspace_,

     1,


        &reshape_164_0_dim_0,

        &reshape_164_0_dim_1,

        &encoder_layers_10_mlp_fc1_weight_dim_0,


        &encoder_layers_10_mlp_fc2_weight_dim_0,

        &encoder_layers_10_mlp_fc2_weight_dim_1,


        &reshape_164_0_dim_0,

        &reshape_164_0_dim_1,

        &encoder_layers_10_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_164_0_dim_0;

        M *= reshape_164_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_10_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_169_0, gemm_rcr_bias_add_167_0, encoder_layers_11_ln1_weight, encoder_layers_11_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_169_0,
        encoder_layers_11_mha_qkv_weight,

        encoder_layers_11_mha_qkv_bias,

        reshape_171_0,
        global_workspace_,
        1,

        &reshape_169_0_dim_0,

        &reshape_169_0_dim_1,


        &encoder_layers_11_mha_qkv_weight_dim_0,

        &encoder_layers_11_mha_qkv_weight_dim_1,


        &reshape_169_0_dim_0,

        &encoder_layers_11_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_172_0,
      split_172_1,
      split_172_2
    };


      int64_t *split_172_0_shape[] = {
        &split_172_0_dim_0, &reshape_171_0_dim_1, &reshape_171_0_dim_2, &reshape_171_0_dim_3, &reshape_171_0_dim_4
      };
      int64_t *split_172_1_shape[] = {
        &split_172_1_dim_0, &reshape_171_0_dim_1, &reshape_171_0_dim_2, &reshape_171_0_dim_3, &reshape_171_0_dim_4
      };
      int64_t *split_172_2_shape[] = {
        &split_172_2_dim_0, &reshape_171_0_dim_1, &reshape_171_0_dim_2, &reshape_171_0_dim_3, &reshape_171_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_172_0_shape, split_172_1_shape, split_172_2_shape
    };

    const int64_t reshape_171_0_shape[] = {
      reshape_171_0_dim_0, reshape_171_0_dim_1, reshape_171_0_dim_2, reshape_171_0_dim_3, reshape_171_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_171_0,
        reshape_171_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_172_0_dim_0,
        &reshape_171_0_dim_1,
        &reshape_171_0_dim_2,
        &reshape_171_0_dim_3,
        &reshape_171_0_dim_4,
        &reshape_173_0_dim_0,
        &reshape_173_0_dim_1,
        &reshape_173_0_dim_2,
        &reshape_173_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_172_1_dim_0,
        &reshape_171_0_dim_1,
        &reshape_171_0_dim_2,
        &reshape_171_0_dim_3,
        &reshape_171_0_dim_4,
        &reshape_174_0_dim_0,
        &reshape_174_0_dim_1,
        &reshape_174_0_dim_2,
        &reshape_174_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_172_2_dim_0,
        &reshape_171_0_dim_1,
        &reshape_171_0_dim_2,
        &reshape_171_0_dim_3,
        &reshape_171_0_dim_4,
        &reshape_175_0_dim_0,
        &reshape_175_0_dim_1,
        &reshape_175_0_dim_2,
        &reshape_175_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_176_0,
        reshape_173_0, reshape_174_0, reshape_175_0,
        &reshape_173_0_dim_0,
        &reshape_175_0_dim_2,
        &reshape_173_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_176_0,
        encoder_layers_11_mha_proj_weight,
        encoder_layers_11_mha_proj_bias,
        gemm_rcr_bias_add_167_0,

        reshape_179_0,
        global_workspace_,

     1,


        &reshape_177_0_dim_0,

        &reshape_177_0_dim_1,


        &encoder_layers_11_mha_proj_weight_dim_0,

        &encoder_layers_11_mha_proj_weight_dim_1,


        &reshape_177_0_dim_0,

        &encoder_layers_11_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_179_0_dim_0;

        M *= reshape_179_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_179_0_dim_2;

    
        layernorm_15(
           layernorm_180_0, reshape_179_0, encoder_layers_11_ln2_weight, encoder_layers_11_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_180_0,
        encoder_layers_11_mlp_fc1_weight,

        encoder_layers_11_mlp_fc1_bias,

        gemm_rcr_bias_gelu_181_0,
        global_workspace_,
        1,

        &reshape_179_0_dim_0,

        &reshape_179_0_dim_1,

        &reshape_179_0_dim_2,


        &encoder_layers_11_mlp_fc1_weight_dim_0,

        &encoder_layers_11_mlp_fc1_weight_dim_1,


        &reshape_179_0_dim_0,

        &reshape_179_0_dim_1,

        &encoder_layers_11_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_181_0,
        encoder_layers_11_mlp_fc2_weight,
        encoder_layers_11_mlp_fc2_bias,
        reshape_179_0,

        gemm_rcr_bias_add_182_0,
        global_workspace_,

     1,


        &reshape_179_0_dim_0,

        &reshape_179_0_dim_1,

        &encoder_layers_11_mlp_fc1_weight_dim_0,


        &encoder_layers_11_mlp_fc2_weight_dim_0,

        &encoder_layers_11_mlp_fc2_weight_dim_1,


        &reshape_179_0_dim_0,

        &reshape_179_0_dim_1,

        &encoder_layers_11_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_179_0_dim_0;

        M *= reshape_179_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_11_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_184_0, gemm_rcr_bias_add_182_0, encoder_layers_12_ln1_weight, encoder_layers_12_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_184_0,
        encoder_layers_12_mha_qkv_weight,

        encoder_layers_12_mha_qkv_bias,

        reshape_186_0,
        global_workspace_,
        1,

        &reshape_184_0_dim_0,

        &reshape_184_0_dim_1,


        &encoder_layers_12_mha_qkv_weight_dim_0,

        &encoder_layers_12_mha_qkv_weight_dim_1,


        &reshape_184_0_dim_0,

        &encoder_layers_12_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_187_2,
      split_187_0,
      split_187_1
    };


      int64_t *split_187_2_shape[] = {
        &split_187_2_dim_0, &reshape_186_0_dim_1, &reshape_186_0_dim_2, &reshape_186_0_dim_3, &reshape_186_0_dim_4
      };
      int64_t *split_187_0_shape[] = {
        &split_187_0_dim_0, &reshape_186_0_dim_1, &reshape_186_0_dim_2, &reshape_186_0_dim_3, &reshape_186_0_dim_4
      };
      int64_t *split_187_1_shape[] = {
        &split_187_1_dim_0, &reshape_186_0_dim_1, &reshape_186_0_dim_2, &reshape_186_0_dim_3, &reshape_186_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_187_2_shape, split_187_0_shape, split_187_1_shape
    };

    const int64_t reshape_186_0_shape[] = {
      reshape_186_0_dim_0, reshape_186_0_dim_1, reshape_186_0_dim_2, reshape_186_0_dim_3, reshape_186_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_186_0,
        reshape_186_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_187_0_dim_0,
        &reshape_186_0_dim_1,
        &reshape_186_0_dim_2,
        &reshape_186_0_dim_3,
        &reshape_186_0_dim_4,
        &reshape_188_0_dim_0,
        &reshape_188_0_dim_1,
        &reshape_188_0_dim_2,
        &reshape_188_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_187_1_dim_0,
        &reshape_186_0_dim_1,
        &reshape_186_0_dim_2,
        &reshape_186_0_dim_3,
        &reshape_186_0_dim_4,
        &reshape_189_0_dim_0,
        &reshape_189_0_dim_1,
        &reshape_189_0_dim_2,
        &reshape_189_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_187_2_dim_0,
        &reshape_186_0_dim_1,
        &reshape_186_0_dim_2,
        &reshape_186_0_dim_3,
        &reshape_186_0_dim_4,
        &reshape_190_0_dim_0,
        &reshape_190_0_dim_1,
        &reshape_190_0_dim_2,
        &reshape_190_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_191_0,
        reshape_190_0, reshape_188_0, reshape_189_0,
        &reshape_190_0_dim_0,
        &reshape_189_0_dim_2,
        &reshape_190_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_191_0,
        encoder_layers_12_mha_proj_weight,
        encoder_layers_12_mha_proj_bias,
        gemm_rcr_bias_add_182_0,

        reshape_194_0,
        global_workspace_,

     1,


        &reshape_192_0_dim_0,

        &reshape_192_0_dim_1,


        &encoder_layers_12_mha_proj_weight_dim_0,

        &encoder_layers_12_mha_proj_weight_dim_1,


        &reshape_192_0_dim_0,

        &encoder_layers_12_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_194_0_dim_0;

        M *= reshape_194_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_194_0_dim_2;

    
        layernorm_15(
           layernorm_195_0, reshape_194_0, encoder_layers_12_ln2_weight, encoder_layers_12_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_195_0,
        encoder_layers_12_mlp_fc1_weight,

        encoder_layers_12_mlp_fc1_bias,

        gemm_rcr_bias_gelu_196_0,
        global_workspace_,
        1,

        &reshape_194_0_dim_0,

        &reshape_194_0_dim_1,

        &reshape_194_0_dim_2,


        &encoder_layers_12_mlp_fc1_weight_dim_0,

        &encoder_layers_12_mlp_fc1_weight_dim_1,


        &reshape_194_0_dim_0,

        &reshape_194_0_dim_1,

        &encoder_layers_12_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_196_0,
        encoder_layers_12_mlp_fc2_weight,
        encoder_layers_12_mlp_fc2_bias,
        reshape_194_0,

        gemm_rcr_bias_add_197_0,
        global_workspace_,

     1,


        &reshape_194_0_dim_0,

        &reshape_194_0_dim_1,

        &encoder_layers_12_mlp_fc1_weight_dim_0,


        &encoder_layers_12_mlp_fc2_weight_dim_0,

        &encoder_layers_12_mlp_fc2_weight_dim_1,


        &reshape_194_0_dim_0,

        &reshape_194_0_dim_1,

        &encoder_layers_12_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_194_0_dim_0;

        M *= reshape_194_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_12_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_199_0, gemm_rcr_bias_add_197_0, encoder_layers_13_ln1_weight, encoder_layers_13_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_199_0,
        encoder_layers_13_mha_qkv_weight,

        encoder_layers_13_mha_qkv_bias,

        reshape_201_0,
        global_workspace_,
        1,

        &reshape_199_0_dim_0,

        &reshape_199_0_dim_1,


        &encoder_layers_13_mha_qkv_weight_dim_0,

        &encoder_layers_13_mha_qkv_weight_dim_1,


        &reshape_199_0_dim_0,

        &encoder_layers_13_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_202_2,
      split_202_1,
      split_202_0
    };


      int64_t *split_202_2_shape[] = {
        &split_202_2_dim_0, &reshape_201_0_dim_1, &reshape_201_0_dim_2, &reshape_201_0_dim_3, &reshape_201_0_dim_4
      };
      int64_t *split_202_1_shape[] = {
        &split_202_1_dim_0, &reshape_201_0_dim_1, &reshape_201_0_dim_2, &reshape_201_0_dim_3, &reshape_201_0_dim_4
      };
      int64_t *split_202_0_shape[] = {
        &split_202_0_dim_0, &reshape_201_0_dim_1, &reshape_201_0_dim_2, &reshape_201_0_dim_3, &reshape_201_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_202_2_shape, split_202_1_shape, split_202_0_shape
    };

    const int64_t reshape_201_0_shape[] = {
      reshape_201_0_dim_0, reshape_201_0_dim_1, reshape_201_0_dim_2, reshape_201_0_dim_3, reshape_201_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_201_0,
        reshape_201_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_202_0_dim_0,
        &reshape_201_0_dim_1,
        &reshape_201_0_dim_2,
        &reshape_201_0_dim_3,
        &reshape_201_0_dim_4,
        &reshape_203_0_dim_0,
        &reshape_203_0_dim_1,
        &reshape_203_0_dim_2,
        &reshape_203_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_202_1_dim_0,
        &reshape_201_0_dim_1,
        &reshape_201_0_dim_2,
        &reshape_201_0_dim_3,
        &reshape_201_0_dim_4,
        &reshape_204_0_dim_0,
        &reshape_204_0_dim_1,
        &reshape_204_0_dim_2,
        &reshape_204_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_202_2_dim_0,
        &reshape_201_0_dim_1,
        &reshape_201_0_dim_2,
        &reshape_201_0_dim_3,
        &reshape_201_0_dim_4,
        &reshape_205_0_dim_0,
        &reshape_205_0_dim_1,
        &reshape_205_0_dim_2,
        &reshape_205_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_206_0,
        reshape_205_0, reshape_204_0, reshape_203_0,
        &reshape_205_0_dim_0,
        &reshape_203_0_dim_2,
        &reshape_205_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_206_0,
        encoder_layers_13_mha_proj_weight,
        encoder_layers_13_mha_proj_bias,
        gemm_rcr_bias_add_197_0,

        reshape_209_0,
        global_workspace_,

     1,


        &reshape_207_0_dim_0,

        &reshape_207_0_dim_1,


        &encoder_layers_13_mha_proj_weight_dim_0,

        &encoder_layers_13_mha_proj_weight_dim_1,


        &reshape_207_0_dim_0,

        &encoder_layers_13_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_209_0_dim_0;

        M *= reshape_209_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_209_0_dim_2;

    
        layernorm_15(
           layernorm_210_0, reshape_209_0, encoder_layers_13_ln2_weight, encoder_layers_13_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_210_0,
        encoder_layers_13_mlp_fc1_weight,

        encoder_layers_13_mlp_fc1_bias,

        gemm_rcr_bias_gelu_211_0,
        global_workspace_,
        1,

        &reshape_209_0_dim_0,

        &reshape_209_0_dim_1,

        &reshape_209_0_dim_2,


        &encoder_layers_13_mlp_fc1_weight_dim_0,

        &encoder_layers_13_mlp_fc1_weight_dim_1,


        &reshape_209_0_dim_0,

        &reshape_209_0_dim_1,

        &encoder_layers_13_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_211_0,
        encoder_layers_13_mlp_fc2_weight,
        encoder_layers_13_mlp_fc2_bias,
        reshape_209_0,

        gemm_rcr_bias_add_212_0,
        global_workspace_,

     1,


        &reshape_209_0_dim_0,

        &reshape_209_0_dim_1,

        &encoder_layers_13_mlp_fc1_weight_dim_0,


        &encoder_layers_13_mlp_fc2_weight_dim_0,

        &encoder_layers_13_mlp_fc2_weight_dim_1,


        &reshape_209_0_dim_0,

        &reshape_209_0_dim_1,

        &encoder_layers_13_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_209_0_dim_0;

        M *= reshape_209_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_13_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_214_0, gemm_rcr_bias_add_212_0, encoder_layers_14_ln1_weight, encoder_layers_14_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_214_0,
        encoder_layers_14_mha_qkv_weight,

        encoder_layers_14_mha_qkv_bias,

        reshape_216_0,
        global_workspace_,
        1,

        &reshape_214_0_dim_0,

        &reshape_214_0_dim_1,


        &encoder_layers_14_mha_qkv_weight_dim_0,

        &encoder_layers_14_mha_qkv_weight_dim_1,


        &reshape_214_0_dim_0,

        &encoder_layers_14_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_217_2,
      split_217_0,
      split_217_1
    };


      int64_t *split_217_2_shape[] = {
        &split_217_2_dim_0, &reshape_216_0_dim_1, &reshape_216_0_dim_2, &reshape_216_0_dim_3, &reshape_216_0_dim_4
      };
      int64_t *split_217_0_shape[] = {
        &split_217_0_dim_0, &reshape_216_0_dim_1, &reshape_216_0_dim_2, &reshape_216_0_dim_3, &reshape_216_0_dim_4
      };
      int64_t *split_217_1_shape[] = {
        &split_217_1_dim_0, &reshape_216_0_dim_1, &reshape_216_0_dim_2, &reshape_216_0_dim_3, &reshape_216_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_217_2_shape, split_217_0_shape, split_217_1_shape
    };

    const int64_t reshape_216_0_shape[] = {
      reshape_216_0_dim_0, reshape_216_0_dim_1, reshape_216_0_dim_2, reshape_216_0_dim_3, reshape_216_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_216_0,
        reshape_216_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_217_0_dim_0,
        &reshape_216_0_dim_1,
        &reshape_216_0_dim_2,
        &reshape_216_0_dim_3,
        &reshape_216_0_dim_4,
        &reshape_218_0_dim_0,
        &reshape_218_0_dim_1,
        &reshape_218_0_dim_2,
        &reshape_218_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_217_1_dim_0,
        &reshape_216_0_dim_1,
        &reshape_216_0_dim_2,
        &reshape_216_0_dim_3,
        &reshape_216_0_dim_4,
        &reshape_219_0_dim_0,
        &reshape_219_0_dim_1,
        &reshape_219_0_dim_2,
        &reshape_219_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_217_2_dim_0,
        &reshape_216_0_dim_1,
        &reshape_216_0_dim_2,
        &reshape_216_0_dim_3,
        &reshape_216_0_dim_4,
        &reshape_220_0_dim_0,
        &reshape_220_0_dim_1,
        &reshape_220_0_dim_2,
        &reshape_220_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_221_0,
        reshape_220_0, reshape_218_0, reshape_219_0,
        &reshape_220_0_dim_0,
        &reshape_219_0_dim_2,
        &reshape_220_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_221_0,
        encoder_layers_14_mha_proj_weight,
        encoder_layers_14_mha_proj_bias,
        gemm_rcr_bias_add_212_0,

        reshape_224_0,
        global_workspace_,

     1,


        &reshape_222_0_dim_0,

        &reshape_222_0_dim_1,


        &encoder_layers_14_mha_proj_weight_dim_0,

        &encoder_layers_14_mha_proj_weight_dim_1,


        &reshape_222_0_dim_0,

        &encoder_layers_14_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_224_0_dim_0;

        M *= reshape_224_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_224_0_dim_2;

    
        layernorm_15(
           layernorm_225_0, reshape_224_0, encoder_layers_14_ln2_weight, encoder_layers_14_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_225_0,
        encoder_layers_14_mlp_fc1_weight,

        encoder_layers_14_mlp_fc1_bias,

        gemm_rcr_bias_gelu_226_0,
        global_workspace_,
        1,

        &reshape_224_0_dim_0,

        &reshape_224_0_dim_1,

        &reshape_224_0_dim_2,


        &encoder_layers_14_mlp_fc1_weight_dim_0,

        &encoder_layers_14_mlp_fc1_weight_dim_1,


        &reshape_224_0_dim_0,

        &reshape_224_0_dim_1,

        &encoder_layers_14_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_226_0,
        encoder_layers_14_mlp_fc2_weight,
        encoder_layers_14_mlp_fc2_bias,
        reshape_224_0,

        gemm_rcr_bias_add_227_0,
        global_workspace_,

     1,


        &reshape_224_0_dim_0,

        &reshape_224_0_dim_1,

        &encoder_layers_14_mlp_fc1_weight_dim_0,


        &encoder_layers_14_mlp_fc2_weight_dim_0,

        &encoder_layers_14_mlp_fc2_weight_dim_1,


        &reshape_224_0_dim_0,

        &reshape_224_0_dim_1,

        &encoder_layers_14_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_224_0_dim_0;

        M *= reshape_224_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_14_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_229_0, gemm_rcr_bias_add_227_0, encoder_layers_15_ln1_weight, encoder_layers_15_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_229_0,
        encoder_layers_15_mha_qkv_weight,

        encoder_layers_15_mha_qkv_bias,

        reshape_231_0,
        global_workspace_,
        1,

        &reshape_229_0_dim_0,

        &reshape_229_0_dim_1,


        &encoder_layers_15_mha_qkv_weight_dim_0,

        &encoder_layers_15_mha_qkv_weight_dim_1,


        &reshape_229_0_dim_0,

        &encoder_layers_15_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_232_0,
      split_232_1,
      split_232_2
    };


      int64_t *split_232_0_shape[] = {
        &split_232_0_dim_0, &reshape_231_0_dim_1, &reshape_231_0_dim_2, &reshape_231_0_dim_3, &reshape_231_0_dim_4
      };
      int64_t *split_232_1_shape[] = {
        &split_232_1_dim_0, &reshape_231_0_dim_1, &reshape_231_0_dim_2, &reshape_231_0_dim_3, &reshape_231_0_dim_4
      };
      int64_t *split_232_2_shape[] = {
        &split_232_2_dim_0, &reshape_231_0_dim_1, &reshape_231_0_dim_2, &reshape_231_0_dim_3, &reshape_231_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_232_0_shape, split_232_1_shape, split_232_2_shape
    };

    const int64_t reshape_231_0_shape[] = {
      reshape_231_0_dim_0, reshape_231_0_dim_1, reshape_231_0_dim_2, reshape_231_0_dim_3, reshape_231_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_231_0,
        reshape_231_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_232_0_dim_0,
        &reshape_231_0_dim_1,
        &reshape_231_0_dim_2,
        &reshape_231_0_dim_3,
        &reshape_231_0_dim_4,
        &reshape_233_0_dim_0,
        &reshape_233_0_dim_1,
        &reshape_233_0_dim_2,
        &reshape_233_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_232_1_dim_0,
        &reshape_231_0_dim_1,
        &reshape_231_0_dim_2,
        &reshape_231_0_dim_3,
        &reshape_231_0_dim_4,
        &reshape_234_0_dim_0,
        &reshape_234_0_dim_1,
        &reshape_234_0_dim_2,
        &reshape_234_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_232_2_dim_0,
        &reshape_231_0_dim_1,
        &reshape_231_0_dim_2,
        &reshape_231_0_dim_3,
        &reshape_231_0_dim_4,
        &reshape_235_0_dim_0,
        &reshape_235_0_dim_1,
        &reshape_235_0_dim_2,
        &reshape_235_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_236_0,
        reshape_233_0, reshape_234_0, reshape_235_0,
        &reshape_233_0_dim_0,
        &reshape_235_0_dim_2,
        &reshape_233_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_236_0,
        encoder_layers_15_mha_proj_weight,
        encoder_layers_15_mha_proj_bias,
        gemm_rcr_bias_add_227_0,

        reshape_239_0,
        global_workspace_,

     1,


        &reshape_237_0_dim_0,

        &reshape_237_0_dim_1,


        &encoder_layers_15_mha_proj_weight_dim_0,

        &encoder_layers_15_mha_proj_weight_dim_1,


        &reshape_237_0_dim_0,

        &encoder_layers_15_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_239_0_dim_0;

        M *= reshape_239_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_239_0_dim_2;

    
        layernorm_15(
           layernorm_240_0, reshape_239_0, encoder_layers_15_ln2_weight, encoder_layers_15_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_240_0,
        encoder_layers_15_mlp_fc1_weight,

        encoder_layers_15_mlp_fc1_bias,

        gemm_rcr_bias_gelu_241_0,
        global_workspace_,
        1,

        &reshape_239_0_dim_0,

        &reshape_239_0_dim_1,

        &reshape_239_0_dim_2,


        &encoder_layers_15_mlp_fc1_weight_dim_0,

        &encoder_layers_15_mlp_fc1_weight_dim_1,


        &reshape_239_0_dim_0,

        &reshape_239_0_dim_1,

        &encoder_layers_15_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_241_0,
        encoder_layers_15_mlp_fc2_weight,
        encoder_layers_15_mlp_fc2_bias,
        reshape_239_0,

        gemm_rcr_bias_add_242_0,
        global_workspace_,

     1,


        &reshape_239_0_dim_0,

        &reshape_239_0_dim_1,

        &encoder_layers_15_mlp_fc1_weight_dim_0,


        &encoder_layers_15_mlp_fc2_weight_dim_0,

        &encoder_layers_15_mlp_fc2_weight_dim_1,


        &reshape_239_0_dim_0,

        &reshape_239_0_dim_1,

        &encoder_layers_15_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_239_0_dim_0;

        M *= reshape_239_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_15_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_244_0, gemm_rcr_bias_add_242_0, encoder_layers_16_ln1_weight, encoder_layers_16_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_244_0,
        encoder_layers_16_mha_qkv_weight,

        encoder_layers_16_mha_qkv_bias,

        reshape_246_0,
        global_workspace_,
        1,

        &reshape_244_0_dim_0,

        &reshape_244_0_dim_1,


        &encoder_layers_16_mha_qkv_weight_dim_0,

        &encoder_layers_16_mha_qkv_weight_dim_1,


        &reshape_244_0_dim_0,

        &encoder_layers_16_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_247_1,
      split_247_0,
      split_247_2
    };


      int64_t *split_247_1_shape[] = {
        &split_247_1_dim_0, &reshape_246_0_dim_1, &reshape_246_0_dim_2, &reshape_246_0_dim_3, &reshape_246_0_dim_4
      };
      int64_t *split_247_0_shape[] = {
        &split_247_0_dim_0, &reshape_246_0_dim_1, &reshape_246_0_dim_2, &reshape_246_0_dim_3, &reshape_246_0_dim_4
      };
      int64_t *split_247_2_shape[] = {
        &split_247_2_dim_0, &reshape_246_0_dim_1, &reshape_246_0_dim_2, &reshape_246_0_dim_3, &reshape_246_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_247_1_shape, split_247_0_shape, split_247_2_shape
    };

    const int64_t reshape_246_0_shape[] = {
      reshape_246_0_dim_0, reshape_246_0_dim_1, reshape_246_0_dim_2, reshape_246_0_dim_3, reshape_246_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_246_0,
        reshape_246_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_247_0_dim_0,
        &reshape_246_0_dim_1,
        &reshape_246_0_dim_2,
        &reshape_246_0_dim_3,
        &reshape_246_0_dim_4,
        &reshape_248_0_dim_0,
        &reshape_248_0_dim_1,
        &reshape_248_0_dim_2,
        &reshape_248_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_247_1_dim_0,
        &reshape_246_0_dim_1,
        &reshape_246_0_dim_2,
        &reshape_246_0_dim_3,
        &reshape_246_0_dim_4,
        &reshape_249_0_dim_0,
        &reshape_249_0_dim_1,
        &reshape_249_0_dim_2,
        &reshape_249_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_247_2_dim_0,
        &reshape_246_0_dim_1,
        &reshape_246_0_dim_2,
        &reshape_246_0_dim_3,
        &reshape_246_0_dim_4,
        &reshape_250_0_dim_0,
        &reshape_250_0_dim_1,
        &reshape_250_0_dim_2,
        &reshape_250_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_251_0,
        reshape_249_0, reshape_248_0, reshape_250_0,
        &reshape_249_0_dim_0,
        &reshape_250_0_dim_2,
        &reshape_249_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_251_0,
        encoder_layers_16_mha_proj_weight,
        encoder_layers_16_mha_proj_bias,
        gemm_rcr_bias_add_242_0,

        reshape_254_0,
        global_workspace_,

     1,


        &reshape_252_0_dim_0,

        &reshape_252_0_dim_1,


        &encoder_layers_16_mha_proj_weight_dim_0,

        &encoder_layers_16_mha_proj_weight_dim_1,


        &reshape_252_0_dim_0,

        &encoder_layers_16_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_254_0_dim_0;

        M *= reshape_254_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_254_0_dim_2;

    
        layernorm_15(
           layernorm_255_0, reshape_254_0, encoder_layers_16_ln2_weight, encoder_layers_16_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_255_0,
        encoder_layers_16_mlp_fc1_weight,

        encoder_layers_16_mlp_fc1_bias,

        gemm_rcr_bias_gelu_256_0,
        global_workspace_,
        1,

        &reshape_254_0_dim_0,

        &reshape_254_0_dim_1,

        &reshape_254_0_dim_2,


        &encoder_layers_16_mlp_fc1_weight_dim_0,

        &encoder_layers_16_mlp_fc1_weight_dim_1,


        &reshape_254_0_dim_0,

        &reshape_254_0_dim_1,

        &encoder_layers_16_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_256_0,
        encoder_layers_16_mlp_fc2_weight,
        encoder_layers_16_mlp_fc2_bias,
        reshape_254_0,

        gemm_rcr_bias_add_257_0,
        global_workspace_,

     1,


        &reshape_254_0_dim_0,

        &reshape_254_0_dim_1,

        &encoder_layers_16_mlp_fc1_weight_dim_0,


        &encoder_layers_16_mlp_fc2_weight_dim_0,

        &encoder_layers_16_mlp_fc2_weight_dim_1,


        &reshape_254_0_dim_0,

        &reshape_254_0_dim_1,

        &encoder_layers_16_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_254_0_dim_0;

        M *= reshape_254_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_16_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_259_0, gemm_rcr_bias_add_257_0, encoder_layers_17_ln1_weight, encoder_layers_17_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_259_0,
        encoder_layers_17_mha_qkv_weight,

        encoder_layers_17_mha_qkv_bias,

        reshape_261_0,
        global_workspace_,
        1,

        &reshape_259_0_dim_0,

        &reshape_259_0_dim_1,


        &encoder_layers_17_mha_qkv_weight_dim_0,

        &encoder_layers_17_mha_qkv_weight_dim_1,


        &reshape_259_0_dim_0,

        &encoder_layers_17_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_262_0,
      split_262_1,
      split_262_2
    };


      int64_t *split_262_0_shape[] = {
        &split_262_0_dim_0, &reshape_261_0_dim_1, &reshape_261_0_dim_2, &reshape_261_0_dim_3, &reshape_261_0_dim_4
      };
      int64_t *split_262_1_shape[] = {
        &split_262_1_dim_0, &reshape_261_0_dim_1, &reshape_261_0_dim_2, &reshape_261_0_dim_3, &reshape_261_0_dim_4
      };
      int64_t *split_262_2_shape[] = {
        &split_262_2_dim_0, &reshape_261_0_dim_1, &reshape_261_0_dim_2, &reshape_261_0_dim_3, &reshape_261_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_262_0_shape, split_262_1_shape, split_262_2_shape
    };

    const int64_t reshape_261_0_shape[] = {
      reshape_261_0_dim_0, reshape_261_0_dim_1, reshape_261_0_dim_2, reshape_261_0_dim_3, reshape_261_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_261_0,
        reshape_261_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_262_0_dim_0,
        &reshape_261_0_dim_1,
        &reshape_261_0_dim_2,
        &reshape_261_0_dim_3,
        &reshape_261_0_dim_4,
        &reshape_263_0_dim_0,
        &reshape_263_0_dim_1,
        &reshape_263_0_dim_2,
        &reshape_263_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_262_1_dim_0,
        &reshape_261_0_dim_1,
        &reshape_261_0_dim_2,
        &reshape_261_0_dim_3,
        &reshape_261_0_dim_4,
        &reshape_264_0_dim_0,
        &reshape_264_0_dim_1,
        &reshape_264_0_dim_2,
        &reshape_264_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_262_2_dim_0,
        &reshape_261_0_dim_1,
        &reshape_261_0_dim_2,
        &reshape_261_0_dim_3,
        &reshape_261_0_dim_4,
        &reshape_265_0_dim_0,
        &reshape_265_0_dim_1,
        &reshape_265_0_dim_2,
        &reshape_265_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_266_0,
        reshape_263_0, reshape_264_0, reshape_265_0,
        &reshape_263_0_dim_0,
        &reshape_265_0_dim_2,
        &reshape_263_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_266_0,
        encoder_layers_17_mha_proj_weight,
        encoder_layers_17_mha_proj_bias,
        gemm_rcr_bias_add_257_0,

        reshape_269_0,
        global_workspace_,

     1,


        &reshape_267_0_dim_0,

        &reshape_267_0_dim_1,


        &encoder_layers_17_mha_proj_weight_dim_0,

        &encoder_layers_17_mha_proj_weight_dim_1,


        &reshape_267_0_dim_0,

        &encoder_layers_17_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_269_0_dim_0;

        M *= reshape_269_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_269_0_dim_2;

    
        layernorm_15(
           layernorm_270_0, reshape_269_0, encoder_layers_17_ln2_weight, encoder_layers_17_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_270_0,
        encoder_layers_17_mlp_fc1_weight,

        encoder_layers_17_mlp_fc1_bias,

        gemm_rcr_bias_gelu_271_0,
        global_workspace_,
        1,

        &reshape_269_0_dim_0,

        &reshape_269_0_dim_1,

        &reshape_269_0_dim_2,


        &encoder_layers_17_mlp_fc1_weight_dim_0,

        &encoder_layers_17_mlp_fc1_weight_dim_1,


        &reshape_269_0_dim_0,

        &reshape_269_0_dim_1,

        &encoder_layers_17_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_271_0,
        encoder_layers_17_mlp_fc2_weight,
        encoder_layers_17_mlp_fc2_bias,
        reshape_269_0,

        gemm_rcr_bias_add_272_0,
        global_workspace_,

     1,


        &reshape_269_0_dim_0,

        &reshape_269_0_dim_1,

        &encoder_layers_17_mlp_fc1_weight_dim_0,


        &encoder_layers_17_mlp_fc2_weight_dim_0,

        &encoder_layers_17_mlp_fc2_weight_dim_1,


        &reshape_269_0_dim_0,

        &reshape_269_0_dim_1,

        &encoder_layers_17_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_269_0_dim_0;

        M *= reshape_269_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_17_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_274_0, gemm_rcr_bias_add_272_0, encoder_layers_18_ln1_weight, encoder_layers_18_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_274_0,
        encoder_layers_18_mha_qkv_weight,

        encoder_layers_18_mha_qkv_bias,

        reshape_276_0,
        global_workspace_,
        1,

        &reshape_274_0_dim_0,

        &reshape_274_0_dim_1,


        &encoder_layers_18_mha_qkv_weight_dim_0,

        &encoder_layers_18_mha_qkv_weight_dim_1,


        &reshape_274_0_dim_0,

        &encoder_layers_18_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_277_1,
      split_277_0,
      split_277_2
    };


      int64_t *split_277_1_shape[] = {
        &split_277_1_dim_0, &reshape_276_0_dim_1, &reshape_276_0_dim_2, &reshape_276_0_dim_3, &reshape_276_0_dim_4
      };
      int64_t *split_277_0_shape[] = {
        &split_277_0_dim_0, &reshape_276_0_dim_1, &reshape_276_0_dim_2, &reshape_276_0_dim_3, &reshape_276_0_dim_4
      };
      int64_t *split_277_2_shape[] = {
        &split_277_2_dim_0, &reshape_276_0_dim_1, &reshape_276_0_dim_2, &reshape_276_0_dim_3, &reshape_276_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_277_1_shape, split_277_0_shape, split_277_2_shape
    };

    const int64_t reshape_276_0_shape[] = {
      reshape_276_0_dim_0, reshape_276_0_dim_1, reshape_276_0_dim_2, reshape_276_0_dim_3, reshape_276_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_276_0,
        reshape_276_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_277_0_dim_0,
        &reshape_276_0_dim_1,
        &reshape_276_0_dim_2,
        &reshape_276_0_dim_3,
        &reshape_276_0_dim_4,
        &reshape_278_0_dim_0,
        &reshape_278_0_dim_1,
        &reshape_278_0_dim_2,
        &reshape_278_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_277_1_dim_0,
        &reshape_276_0_dim_1,
        &reshape_276_0_dim_2,
        &reshape_276_0_dim_3,
        &reshape_276_0_dim_4,
        &reshape_279_0_dim_0,
        &reshape_279_0_dim_1,
        &reshape_279_0_dim_2,
        &reshape_279_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_277_2_dim_0,
        &reshape_276_0_dim_1,
        &reshape_276_0_dim_2,
        &reshape_276_0_dim_3,
        &reshape_276_0_dim_4,
        &reshape_280_0_dim_0,
        &reshape_280_0_dim_1,
        &reshape_280_0_dim_2,
        &reshape_280_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_281_0,
        reshape_279_0, reshape_278_0, reshape_280_0,
        &reshape_279_0_dim_0,
        &reshape_280_0_dim_2,
        &reshape_279_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_281_0,
        encoder_layers_18_mha_proj_weight,
        encoder_layers_18_mha_proj_bias,
        gemm_rcr_bias_add_272_0,

        reshape_284_0,
        global_workspace_,

     1,


        &reshape_282_0_dim_0,

        &reshape_282_0_dim_1,


        &encoder_layers_18_mha_proj_weight_dim_0,

        &encoder_layers_18_mha_proj_weight_dim_1,


        &reshape_282_0_dim_0,

        &encoder_layers_18_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_284_0_dim_0;

        M *= reshape_284_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_284_0_dim_2;

    
        layernorm_15(
           layernorm_285_0, reshape_284_0, encoder_layers_18_ln2_weight, encoder_layers_18_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_285_0,
        encoder_layers_18_mlp_fc1_weight,

        encoder_layers_18_mlp_fc1_bias,

        gemm_rcr_bias_gelu_286_0,
        global_workspace_,
        1,

        &reshape_284_0_dim_0,

        &reshape_284_0_dim_1,

        &reshape_284_0_dim_2,


        &encoder_layers_18_mlp_fc1_weight_dim_0,

        &encoder_layers_18_mlp_fc1_weight_dim_1,


        &reshape_284_0_dim_0,

        &reshape_284_0_dim_1,

        &encoder_layers_18_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_286_0,
        encoder_layers_18_mlp_fc2_weight,
        encoder_layers_18_mlp_fc2_bias,
        reshape_284_0,

        gemm_rcr_bias_add_287_0,
        global_workspace_,

     1,


        &reshape_284_0_dim_0,

        &reshape_284_0_dim_1,

        &encoder_layers_18_mlp_fc1_weight_dim_0,


        &encoder_layers_18_mlp_fc2_weight_dim_0,

        &encoder_layers_18_mlp_fc2_weight_dim_1,


        &reshape_284_0_dim_0,

        &reshape_284_0_dim_1,

        &encoder_layers_18_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_284_0_dim_0;

        M *= reshape_284_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_18_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_289_0, gemm_rcr_bias_add_287_0, encoder_layers_19_ln1_weight, encoder_layers_19_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_289_0,
        encoder_layers_19_mha_qkv_weight,

        encoder_layers_19_mha_qkv_bias,

        reshape_291_0,
        global_workspace_,
        1,

        &reshape_289_0_dim_0,

        &reshape_289_0_dim_1,


        &encoder_layers_19_mha_qkv_weight_dim_0,

        &encoder_layers_19_mha_qkv_weight_dim_1,


        &reshape_289_0_dim_0,

        &encoder_layers_19_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_292_0,
      split_292_1,
      split_292_2
    };


      int64_t *split_292_0_shape[] = {
        &split_292_0_dim_0, &reshape_291_0_dim_1, &reshape_291_0_dim_2, &reshape_291_0_dim_3, &reshape_291_0_dim_4
      };
      int64_t *split_292_1_shape[] = {
        &split_292_1_dim_0, &reshape_291_0_dim_1, &reshape_291_0_dim_2, &reshape_291_0_dim_3, &reshape_291_0_dim_4
      };
      int64_t *split_292_2_shape[] = {
        &split_292_2_dim_0, &reshape_291_0_dim_1, &reshape_291_0_dim_2, &reshape_291_0_dim_3, &reshape_291_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_292_0_shape, split_292_1_shape, split_292_2_shape
    };

    const int64_t reshape_291_0_shape[] = {
      reshape_291_0_dim_0, reshape_291_0_dim_1, reshape_291_0_dim_2, reshape_291_0_dim_3, reshape_291_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_291_0,
        reshape_291_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_292_0_dim_0,
        &reshape_291_0_dim_1,
        &reshape_291_0_dim_2,
        &reshape_291_0_dim_3,
        &reshape_291_0_dim_4,
        &reshape_293_0_dim_0,
        &reshape_293_0_dim_1,
        &reshape_293_0_dim_2,
        &reshape_293_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_292_1_dim_0,
        &reshape_291_0_dim_1,
        &reshape_291_0_dim_2,
        &reshape_291_0_dim_3,
        &reshape_291_0_dim_4,
        &reshape_294_0_dim_0,
        &reshape_294_0_dim_1,
        &reshape_294_0_dim_2,
        &reshape_294_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_292_2_dim_0,
        &reshape_291_0_dim_1,
        &reshape_291_0_dim_2,
        &reshape_291_0_dim_3,
        &reshape_291_0_dim_4,
        &reshape_295_0_dim_0,
        &reshape_295_0_dim_1,
        &reshape_295_0_dim_2,
        &reshape_295_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_296_0,
        reshape_293_0, reshape_294_0, reshape_295_0,
        &reshape_293_0_dim_0,
        &reshape_295_0_dim_2,
        &reshape_293_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_296_0,
        encoder_layers_19_mha_proj_weight,
        encoder_layers_19_mha_proj_bias,
        gemm_rcr_bias_add_287_0,

        reshape_299_0,
        global_workspace_,

     1,


        &reshape_297_0_dim_0,

        &reshape_297_0_dim_1,


        &encoder_layers_19_mha_proj_weight_dim_0,

        &encoder_layers_19_mha_proj_weight_dim_1,


        &reshape_297_0_dim_0,

        &encoder_layers_19_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_299_0_dim_0;

        M *= reshape_299_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_299_0_dim_2;

    
        layernorm_15(
           layernorm_300_0, reshape_299_0, encoder_layers_19_ln2_weight, encoder_layers_19_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_300_0,
        encoder_layers_19_mlp_fc1_weight,

        encoder_layers_19_mlp_fc1_bias,

        gemm_rcr_bias_gelu_301_0,
        global_workspace_,
        1,

        &reshape_299_0_dim_0,

        &reshape_299_0_dim_1,

        &reshape_299_0_dim_2,


        &encoder_layers_19_mlp_fc1_weight_dim_0,

        &encoder_layers_19_mlp_fc1_weight_dim_1,


        &reshape_299_0_dim_0,

        &reshape_299_0_dim_1,

        &encoder_layers_19_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_301_0,
        encoder_layers_19_mlp_fc2_weight,
        encoder_layers_19_mlp_fc2_bias,
        reshape_299_0,

        gemm_rcr_bias_add_302_0,
        global_workspace_,

     1,


        &reshape_299_0_dim_0,

        &reshape_299_0_dim_1,

        &encoder_layers_19_mlp_fc1_weight_dim_0,


        &encoder_layers_19_mlp_fc2_weight_dim_0,

        &encoder_layers_19_mlp_fc2_weight_dim_1,


        &reshape_299_0_dim_0,

        &reshape_299_0_dim_1,

        &encoder_layers_19_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_299_0_dim_0;

        M *= reshape_299_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_19_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_304_0, gemm_rcr_bias_add_302_0, encoder_layers_20_ln1_weight, encoder_layers_20_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_304_0,
        encoder_layers_20_mha_qkv_weight,

        encoder_layers_20_mha_qkv_bias,

        reshape_306_0,
        global_workspace_,
        1,

        &reshape_304_0_dim_0,

        &reshape_304_0_dim_1,


        &encoder_layers_20_mha_qkv_weight_dim_0,

        &encoder_layers_20_mha_qkv_weight_dim_1,


        &reshape_304_0_dim_0,

        &encoder_layers_20_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_307_2,
      split_307_1,
      split_307_0
    };


      int64_t *split_307_2_shape[] = {
        &split_307_2_dim_0, &reshape_306_0_dim_1, &reshape_306_0_dim_2, &reshape_306_0_dim_3, &reshape_306_0_dim_4
      };
      int64_t *split_307_1_shape[] = {
        &split_307_1_dim_0, &reshape_306_0_dim_1, &reshape_306_0_dim_2, &reshape_306_0_dim_3, &reshape_306_0_dim_4
      };
      int64_t *split_307_0_shape[] = {
        &split_307_0_dim_0, &reshape_306_0_dim_1, &reshape_306_0_dim_2, &reshape_306_0_dim_3, &reshape_306_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_307_2_shape, split_307_1_shape, split_307_0_shape
    };

    const int64_t reshape_306_0_shape[] = {
      reshape_306_0_dim_0, reshape_306_0_dim_1, reshape_306_0_dim_2, reshape_306_0_dim_3, reshape_306_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_306_0,
        reshape_306_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_307_0_dim_0,
        &reshape_306_0_dim_1,
        &reshape_306_0_dim_2,
        &reshape_306_0_dim_3,
        &reshape_306_0_dim_4,
        &reshape_308_0_dim_0,
        &reshape_308_0_dim_1,
        &reshape_308_0_dim_2,
        &reshape_308_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_307_1_dim_0,
        &reshape_306_0_dim_1,
        &reshape_306_0_dim_2,
        &reshape_306_0_dim_3,
        &reshape_306_0_dim_4,
        &reshape_309_0_dim_0,
        &reshape_309_0_dim_1,
        &reshape_309_0_dim_2,
        &reshape_309_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_307_2_dim_0,
        &reshape_306_0_dim_1,
        &reshape_306_0_dim_2,
        &reshape_306_0_dim_3,
        &reshape_306_0_dim_4,
        &reshape_310_0_dim_0,
        &reshape_310_0_dim_1,
        &reshape_310_0_dim_2,
        &reshape_310_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_311_0,
        reshape_310_0, reshape_309_0, reshape_308_0,
        &reshape_310_0_dim_0,
        &reshape_308_0_dim_2,
        &reshape_310_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_311_0,
        encoder_layers_20_mha_proj_weight,
        encoder_layers_20_mha_proj_bias,
        gemm_rcr_bias_add_302_0,

        reshape_314_0,
        global_workspace_,

     1,


        &reshape_312_0_dim_0,

        &reshape_312_0_dim_1,


        &encoder_layers_20_mha_proj_weight_dim_0,

        &encoder_layers_20_mha_proj_weight_dim_1,


        &reshape_312_0_dim_0,

        &encoder_layers_20_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_314_0_dim_0;

        M *= reshape_314_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_314_0_dim_2;

    
        layernorm_15(
           layernorm_315_0, reshape_314_0, encoder_layers_20_ln2_weight, encoder_layers_20_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_315_0,
        encoder_layers_20_mlp_fc1_weight,

        encoder_layers_20_mlp_fc1_bias,

        gemm_rcr_bias_gelu_316_0,
        global_workspace_,
        1,

        &reshape_314_0_dim_0,

        &reshape_314_0_dim_1,

        &reshape_314_0_dim_2,


        &encoder_layers_20_mlp_fc1_weight_dim_0,

        &encoder_layers_20_mlp_fc1_weight_dim_1,


        &reshape_314_0_dim_0,

        &reshape_314_0_dim_1,

        &encoder_layers_20_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_316_0,
        encoder_layers_20_mlp_fc2_weight,
        encoder_layers_20_mlp_fc2_bias,
        reshape_314_0,

        gemm_rcr_bias_add_317_0,
        global_workspace_,

     1,


        &reshape_314_0_dim_0,

        &reshape_314_0_dim_1,

        &encoder_layers_20_mlp_fc1_weight_dim_0,


        &encoder_layers_20_mlp_fc2_weight_dim_0,

        &encoder_layers_20_mlp_fc2_weight_dim_1,


        &reshape_314_0_dim_0,

        &reshape_314_0_dim_1,

        &encoder_layers_20_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_314_0_dim_0;

        M *= reshape_314_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_20_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_319_0, gemm_rcr_bias_add_317_0, encoder_layers_21_ln1_weight, encoder_layers_21_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_319_0,
        encoder_layers_21_mha_qkv_weight,

        encoder_layers_21_mha_qkv_bias,

        reshape_321_0,
        global_workspace_,
        1,

        &reshape_319_0_dim_0,

        &reshape_319_0_dim_1,


        &encoder_layers_21_mha_qkv_weight_dim_0,

        &encoder_layers_21_mha_qkv_weight_dim_1,


        &reshape_319_0_dim_0,

        &encoder_layers_21_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_322_1,
      split_322_2,
      split_322_0
    };


      int64_t *split_322_1_shape[] = {
        &split_322_1_dim_0, &reshape_321_0_dim_1, &reshape_321_0_dim_2, &reshape_321_0_dim_3, &reshape_321_0_dim_4
      };
      int64_t *split_322_2_shape[] = {
        &split_322_2_dim_0, &reshape_321_0_dim_1, &reshape_321_0_dim_2, &reshape_321_0_dim_3, &reshape_321_0_dim_4
      };
      int64_t *split_322_0_shape[] = {
        &split_322_0_dim_0, &reshape_321_0_dim_1, &reshape_321_0_dim_2, &reshape_321_0_dim_3, &reshape_321_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_322_1_shape, split_322_2_shape, split_322_0_shape
    };

    const int64_t reshape_321_0_shape[] = {
      reshape_321_0_dim_0, reshape_321_0_dim_1, reshape_321_0_dim_2, reshape_321_0_dim_3, reshape_321_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_321_0,
        reshape_321_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_322_0_dim_0,
        &reshape_321_0_dim_1,
        &reshape_321_0_dim_2,
        &reshape_321_0_dim_3,
        &reshape_321_0_dim_4,
        &reshape_323_0_dim_0,
        &reshape_323_0_dim_1,
        &reshape_323_0_dim_2,
        &reshape_323_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_322_1_dim_0,
        &reshape_321_0_dim_1,
        &reshape_321_0_dim_2,
        &reshape_321_0_dim_3,
        &reshape_321_0_dim_4,
        &reshape_324_0_dim_0,
        &reshape_324_0_dim_1,
        &reshape_324_0_dim_2,
        &reshape_324_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_322_2_dim_0,
        &reshape_321_0_dim_1,
        &reshape_321_0_dim_2,
        &reshape_321_0_dim_3,
        &reshape_321_0_dim_4,
        &reshape_325_0_dim_0,
        &reshape_325_0_dim_1,
        &reshape_325_0_dim_2,
        &reshape_325_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_326_0,
        reshape_324_0, reshape_325_0, reshape_323_0,
        &reshape_324_0_dim_0,
        &reshape_323_0_dim_2,
        &reshape_324_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_326_0,
        encoder_layers_21_mha_proj_weight,
        encoder_layers_21_mha_proj_bias,
        gemm_rcr_bias_add_317_0,

        reshape_329_0,
        global_workspace_,

     1,


        &reshape_327_0_dim_0,

        &reshape_327_0_dim_1,


        &encoder_layers_21_mha_proj_weight_dim_0,

        &encoder_layers_21_mha_proj_weight_dim_1,


        &reshape_327_0_dim_0,

        &encoder_layers_21_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_329_0_dim_0;

        M *= reshape_329_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_329_0_dim_2;

    
        layernorm_15(
           layernorm_330_0, reshape_329_0, encoder_layers_21_ln2_weight, encoder_layers_21_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_330_0,
        encoder_layers_21_mlp_fc1_weight,

        encoder_layers_21_mlp_fc1_bias,

        gemm_rcr_bias_gelu_331_0,
        global_workspace_,
        1,

        &reshape_329_0_dim_0,

        &reshape_329_0_dim_1,

        &reshape_329_0_dim_2,


        &encoder_layers_21_mlp_fc1_weight_dim_0,

        &encoder_layers_21_mlp_fc1_weight_dim_1,


        &reshape_329_0_dim_0,

        &reshape_329_0_dim_1,

        &encoder_layers_21_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_331_0,
        encoder_layers_21_mlp_fc2_weight,
        encoder_layers_21_mlp_fc2_bias,
        reshape_329_0,

        gemm_rcr_bias_add_332_0,
        global_workspace_,

     1,


        &reshape_329_0_dim_0,

        &reshape_329_0_dim_1,

        &encoder_layers_21_mlp_fc1_weight_dim_0,


        &encoder_layers_21_mlp_fc2_weight_dim_0,

        &encoder_layers_21_mlp_fc2_weight_dim_1,


        &reshape_329_0_dim_0,

        &reshape_329_0_dim_1,

        &encoder_layers_21_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_329_0_dim_0;

        M *= reshape_329_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_21_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_334_0, gemm_rcr_bias_add_332_0, encoder_layers_22_ln1_weight, encoder_layers_22_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_334_0,
        encoder_layers_22_mha_qkv_weight,

        encoder_layers_22_mha_qkv_bias,

        reshape_336_0,
        global_workspace_,
        1,

        &reshape_334_0_dim_0,

        &reshape_334_0_dim_1,


        &encoder_layers_22_mha_qkv_weight_dim_0,

        &encoder_layers_22_mha_qkv_weight_dim_1,


        &reshape_334_0_dim_0,

        &encoder_layers_22_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_337_0,
      split_337_1,
      split_337_2
    };


      int64_t *split_337_0_shape[] = {
        &split_337_0_dim_0, &reshape_336_0_dim_1, &reshape_336_0_dim_2, &reshape_336_0_dim_3, &reshape_336_0_dim_4
      };
      int64_t *split_337_1_shape[] = {
        &split_337_1_dim_0, &reshape_336_0_dim_1, &reshape_336_0_dim_2, &reshape_336_0_dim_3, &reshape_336_0_dim_4
      };
      int64_t *split_337_2_shape[] = {
        &split_337_2_dim_0, &reshape_336_0_dim_1, &reshape_336_0_dim_2, &reshape_336_0_dim_3, &reshape_336_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_337_0_shape, split_337_1_shape, split_337_2_shape
    };

    const int64_t reshape_336_0_shape[] = {
      reshape_336_0_dim_0, reshape_336_0_dim_1, reshape_336_0_dim_2, reshape_336_0_dim_3, reshape_336_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_336_0,
        reshape_336_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_337_0_dim_0,
        &reshape_336_0_dim_1,
        &reshape_336_0_dim_2,
        &reshape_336_0_dim_3,
        &reshape_336_0_dim_4,
        &reshape_338_0_dim_0,
        &reshape_338_0_dim_1,
        &reshape_338_0_dim_2,
        &reshape_338_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_337_1_dim_0,
        &reshape_336_0_dim_1,
        &reshape_336_0_dim_2,
        &reshape_336_0_dim_3,
        &reshape_336_0_dim_4,
        &reshape_339_0_dim_0,
        &reshape_339_0_dim_1,
        &reshape_339_0_dim_2,
        &reshape_339_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_337_2_dim_0,
        &reshape_336_0_dim_1,
        &reshape_336_0_dim_2,
        &reshape_336_0_dim_3,
        &reshape_336_0_dim_4,
        &reshape_340_0_dim_0,
        &reshape_340_0_dim_1,
        &reshape_340_0_dim_2,
        &reshape_340_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_341_0,
        reshape_338_0, reshape_339_0, reshape_340_0,
        &reshape_338_0_dim_0,
        &reshape_340_0_dim_2,
        &reshape_338_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_341_0,
        encoder_layers_22_mha_proj_weight,
        encoder_layers_22_mha_proj_bias,
        gemm_rcr_bias_add_332_0,

        reshape_344_0,
        global_workspace_,

     1,


        &reshape_342_0_dim_0,

        &reshape_342_0_dim_1,


        &encoder_layers_22_mha_proj_weight_dim_0,

        &encoder_layers_22_mha_proj_weight_dim_1,


        &reshape_342_0_dim_0,

        &encoder_layers_22_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_344_0_dim_0;

        M *= reshape_344_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_344_0_dim_2;

    
        layernorm_15(
           layernorm_345_0, reshape_344_0, encoder_layers_22_ln2_weight, encoder_layers_22_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_345_0,
        encoder_layers_22_mlp_fc1_weight,

        encoder_layers_22_mlp_fc1_bias,

        gemm_rcr_bias_gelu_346_0,
        global_workspace_,
        1,

        &reshape_344_0_dim_0,

        &reshape_344_0_dim_1,

        &reshape_344_0_dim_2,


        &encoder_layers_22_mlp_fc1_weight_dim_0,

        &encoder_layers_22_mlp_fc1_weight_dim_1,


        &reshape_344_0_dim_0,

        &reshape_344_0_dim_1,

        &encoder_layers_22_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_346_0,
        encoder_layers_22_mlp_fc2_weight,
        encoder_layers_22_mlp_fc2_bias,
        reshape_344_0,

        gemm_rcr_bias_add_347_0,
        global_workspace_,

     1,


        &reshape_344_0_dim_0,

        &reshape_344_0_dim_1,

        &encoder_layers_22_mlp_fc1_weight_dim_0,


        &encoder_layers_22_mlp_fc2_weight_dim_0,

        &encoder_layers_22_mlp_fc2_weight_dim_1,


        &reshape_344_0_dim_0,

        &reshape_344_0_dim_1,

        &encoder_layers_22_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_344_0_dim_0;

        M *= reshape_344_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_22_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_349_0, gemm_rcr_bias_add_347_0, encoder_layers_23_ln1_weight, encoder_layers_23_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_349_0,
        encoder_layers_23_mha_qkv_weight,

        encoder_layers_23_mha_qkv_bias,

        reshape_351_0,
        global_workspace_,
        1,

        &reshape_349_0_dim_0,

        &reshape_349_0_dim_1,


        &encoder_layers_23_mha_qkv_weight_dim_0,

        &encoder_layers_23_mha_qkv_weight_dim_1,


        &reshape_349_0_dim_0,

        &encoder_layers_23_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_352_1,
      split_352_2,
      split_352_0
    };


      int64_t *split_352_1_shape[] = {
        &split_352_1_dim_0, &reshape_351_0_dim_1, &reshape_351_0_dim_2, &reshape_351_0_dim_3, &reshape_351_0_dim_4
      };
      int64_t *split_352_2_shape[] = {
        &split_352_2_dim_0, &reshape_351_0_dim_1, &reshape_351_0_dim_2, &reshape_351_0_dim_3, &reshape_351_0_dim_4
      };
      int64_t *split_352_0_shape[] = {
        &split_352_0_dim_0, &reshape_351_0_dim_1, &reshape_351_0_dim_2, &reshape_351_0_dim_3, &reshape_351_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_352_1_shape, split_352_2_shape, split_352_0_shape
    };

    const int64_t reshape_351_0_shape[] = {
      reshape_351_0_dim_0, reshape_351_0_dim_1, reshape_351_0_dim_2, reshape_351_0_dim_3, reshape_351_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_351_0,
        reshape_351_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_352_0_dim_0,
        &reshape_351_0_dim_1,
        &reshape_351_0_dim_2,
        &reshape_351_0_dim_3,
        &reshape_351_0_dim_4,
        &reshape_353_0_dim_0,
        &reshape_353_0_dim_1,
        &reshape_353_0_dim_2,
        &reshape_353_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_352_1_dim_0,
        &reshape_351_0_dim_1,
        &reshape_351_0_dim_2,
        &reshape_351_0_dim_3,
        &reshape_351_0_dim_4,
        &reshape_354_0_dim_0,
        &reshape_354_0_dim_1,
        &reshape_354_0_dim_2,
        &reshape_354_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_352_2_dim_0,
        &reshape_351_0_dim_1,
        &reshape_351_0_dim_2,
        &reshape_351_0_dim_3,
        &reshape_351_0_dim_4,
        &reshape_355_0_dim_0,
        &reshape_355_0_dim_1,
        &reshape_355_0_dim_2,
        &reshape_355_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_356_0,
        reshape_354_0, reshape_355_0, reshape_353_0,
        &reshape_354_0_dim_0,
        &reshape_353_0_dim_2,
        &reshape_354_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_356_0,
        encoder_layers_23_mha_proj_weight,
        encoder_layers_23_mha_proj_bias,
        gemm_rcr_bias_add_347_0,

        reshape_359_0,
        global_workspace_,

     1,


        &reshape_357_0_dim_0,

        &reshape_357_0_dim_1,


        &encoder_layers_23_mha_proj_weight_dim_0,

        &encoder_layers_23_mha_proj_weight_dim_1,


        &reshape_357_0_dim_0,

        &encoder_layers_23_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_359_0_dim_0;

        M *= reshape_359_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_359_0_dim_2;

    
        layernorm_15(
           layernorm_360_0, reshape_359_0, encoder_layers_23_ln2_weight, encoder_layers_23_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_360_0,
        encoder_layers_23_mlp_fc1_weight,

        encoder_layers_23_mlp_fc1_bias,

        gemm_rcr_bias_gelu_361_0,
        global_workspace_,
        1,

        &reshape_359_0_dim_0,

        &reshape_359_0_dim_1,

        &reshape_359_0_dim_2,


        &encoder_layers_23_mlp_fc1_weight_dim_0,

        &encoder_layers_23_mlp_fc1_weight_dim_1,


        &reshape_359_0_dim_0,

        &reshape_359_0_dim_1,

        &encoder_layers_23_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_361_0,
        encoder_layers_23_mlp_fc2_weight,
        encoder_layers_23_mlp_fc2_bias,
        reshape_359_0,

        gemm_rcr_bias_add_362_0,
        global_workspace_,

     1,


        &reshape_359_0_dim_0,

        &reshape_359_0_dim_1,

        &encoder_layers_23_mlp_fc1_weight_dim_0,


        &encoder_layers_23_mlp_fc2_weight_dim_0,

        &encoder_layers_23_mlp_fc2_weight_dim_1,


        &reshape_359_0_dim_0,

        &reshape_359_0_dim_1,

        &encoder_layers_23_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_359_0_dim_0;

        M *= reshape_359_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_23_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_364_0, gemm_rcr_bias_add_362_0, encoder_layers_24_ln1_weight, encoder_layers_24_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_364_0,
        encoder_layers_24_mha_qkv_weight,

        encoder_layers_24_mha_qkv_bias,

        reshape_366_0,
        global_workspace_,
        1,

        &reshape_364_0_dim_0,

        &reshape_364_0_dim_1,


        &encoder_layers_24_mha_qkv_weight_dim_0,

        &encoder_layers_24_mha_qkv_weight_dim_1,


        &reshape_364_0_dim_0,

        &encoder_layers_24_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_367_0,
      split_367_2,
      split_367_1
    };


      int64_t *split_367_0_shape[] = {
        &split_367_0_dim_0, &reshape_366_0_dim_1, &reshape_366_0_dim_2, &reshape_366_0_dim_3, &reshape_366_0_dim_4
      };
      int64_t *split_367_2_shape[] = {
        &split_367_2_dim_0, &reshape_366_0_dim_1, &reshape_366_0_dim_2, &reshape_366_0_dim_3, &reshape_366_0_dim_4
      };
      int64_t *split_367_1_shape[] = {
        &split_367_1_dim_0, &reshape_366_0_dim_1, &reshape_366_0_dim_2, &reshape_366_0_dim_3, &reshape_366_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_367_0_shape, split_367_2_shape, split_367_1_shape
    };

    const int64_t reshape_366_0_shape[] = {
      reshape_366_0_dim_0, reshape_366_0_dim_1, reshape_366_0_dim_2, reshape_366_0_dim_3, reshape_366_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_366_0,
        reshape_366_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_367_0_dim_0,
        &reshape_366_0_dim_1,
        &reshape_366_0_dim_2,
        &reshape_366_0_dim_3,
        &reshape_366_0_dim_4,
        &reshape_368_0_dim_0,
        &reshape_368_0_dim_1,
        &reshape_368_0_dim_2,
        &reshape_368_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_367_1_dim_0,
        &reshape_366_0_dim_1,
        &reshape_366_0_dim_2,
        &reshape_366_0_dim_3,
        &reshape_366_0_dim_4,
        &reshape_369_0_dim_0,
        &reshape_369_0_dim_1,
        &reshape_369_0_dim_2,
        &reshape_369_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_367_2_dim_0,
        &reshape_366_0_dim_1,
        &reshape_366_0_dim_2,
        &reshape_366_0_dim_3,
        &reshape_366_0_dim_4,
        &reshape_370_0_dim_0,
        &reshape_370_0_dim_1,
        &reshape_370_0_dim_2,
        &reshape_370_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_371_0,
        reshape_368_0, reshape_370_0, reshape_369_0,
        &reshape_368_0_dim_0,
        &reshape_369_0_dim_2,
        &reshape_368_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_371_0,
        encoder_layers_24_mha_proj_weight,
        encoder_layers_24_mha_proj_bias,
        gemm_rcr_bias_add_362_0,

        reshape_374_0,
        global_workspace_,

     1,


        &reshape_372_0_dim_0,

        &reshape_372_0_dim_1,


        &encoder_layers_24_mha_proj_weight_dim_0,

        &encoder_layers_24_mha_proj_weight_dim_1,


        &reshape_372_0_dim_0,

        &encoder_layers_24_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_374_0_dim_0;

        M *= reshape_374_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_374_0_dim_2;

    
        layernorm_15(
           layernorm_375_0, reshape_374_0, encoder_layers_24_ln2_weight, encoder_layers_24_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_375_0,
        encoder_layers_24_mlp_fc1_weight,

        encoder_layers_24_mlp_fc1_bias,

        gemm_rcr_bias_gelu_376_0,
        global_workspace_,
        1,

        &reshape_374_0_dim_0,

        &reshape_374_0_dim_1,

        &reshape_374_0_dim_2,


        &encoder_layers_24_mlp_fc1_weight_dim_0,

        &encoder_layers_24_mlp_fc1_weight_dim_1,


        &reshape_374_0_dim_0,

        &reshape_374_0_dim_1,

        &encoder_layers_24_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_376_0,
        encoder_layers_24_mlp_fc2_weight,
        encoder_layers_24_mlp_fc2_bias,
        reshape_374_0,

        gemm_rcr_bias_add_377_0,
        global_workspace_,

     1,


        &reshape_374_0_dim_0,

        &reshape_374_0_dim_1,

        &encoder_layers_24_mlp_fc1_weight_dim_0,


        &encoder_layers_24_mlp_fc2_weight_dim_0,

        &encoder_layers_24_mlp_fc2_weight_dim_1,


        &reshape_374_0_dim_0,

        &reshape_374_0_dim_1,

        &encoder_layers_24_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_374_0_dim_0;

        M *= reshape_374_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_24_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_379_0, gemm_rcr_bias_add_377_0, encoder_layers_25_ln1_weight, encoder_layers_25_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_379_0,
        encoder_layers_25_mha_qkv_weight,

        encoder_layers_25_mha_qkv_bias,

        reshape_381_0,
        global_workspace_,
        1,

        &reshape_379_0_dim_0,

        &reshape_379_0_dim_1,


        &encoder_layers_25_mha_qkv_weight_dim_0,

        &encoder_layers_25_mha_qkv_weight_dim_1,


        &reshape_379_0_dim_0,

        &encoder_layers_25_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_382_1,
      split_382_0,
      split_382_2
    };


      int64_t *split_382_1_shape[] = {
        &split_382_1_dim_0, &reshape_381_0_dim_1, &reshape_381_0_dim_2, &reshape_381_0_dim_3, &reshape_381_0_dim_4
      };
      int64_t *split_382_0_shape[] = {
        &split_382_0_dim_0, &reshape_381_0_dim_1, &reshape_381_0_dim_2, &reshape_381_0_dim_3, &reshape_381_0_dim_4
      };
      int64_t *split_382_2_shape[] = {
        &split_382_2_dim_0, &reshape_381_0_dim_1, &reshape_381_0_dim_2, &reshape_381_0_dim_3, &reshape_381_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_382_1_shape, split_382_0_shape, split_382_2_shape
    };

    const int64_t reshape_381_0_shape[] = {
      reshape_381_0_dim_0, reshape_381_0_dim_1, reshape_381_0_dim_2, reshape_381_0_dim_3, reshape_381_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_381_0,
        reshape_381_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_382_0_dim_0,
        &reshape_381_0_dim_1,
        &reshape_381_0_dim_2,
        &reshape_381_0_dim_3,
        &reshape_381_0_dim_4,
        &reshape_383_0_dim_0,
        &reshape_383_0_dim_1,
        &reshape_383_0_dim_2,
        &reshape_383_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_382_1_dim_0,
        &reshape_381_0_dim_1,
        &reshape_381_0_dim_2,
        &reshape_381_0_dim_3,
        &reshape_381_0_dim_4,
        &reshape_384_0_dim_0,
        &reshape_384_0_dim_1,
        &reshape_384_0_dim_2,
        &reshape_384_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_382_2_dim_0,
        &reshape_381_0_dim_1,
        &reshape_381_0_dim_2,
        &reshape_381_0_dim_3,
        &reshape_381_0_dim_4,
        &reshape_385_0_dim_0,
        &reshape_385_0_dim_1,
        &reshape_385_0_dim_2,
        &reshape_385_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_386_0,
        reshape_384_0, reshape_383_0, reshape_385_0,
        &reshape_384_0_dim_0,
        &reshape_385_0_dim_2,
        &reshape_384_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_386_0,
        encoder_layers_25_mha_proj_weight,
        encoder_layers_25_mha_proj_bias,
        gemm_rcr_bias_add_377_0,

        reshape_389_0,
        global_workspace_,

     1,


        &reshape_387_0_dim_0,

        &reshape_387_0_dim_1,


        &encoder_layers_25_mha_proj_weight_dim_0,

        &encoder_layers_25_mha_proj_weight_dim_1,


        &reshape_387_0_dim_0,

        &encoder_layers_25_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_389_0_dim_0;

        M *= reshape_389_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_389_0_dim_2;

    
        layernorm_15(
           layernorm_390_0, reshape_389_0, encoder_layers_25_ln2_weight, encoder_layers_25_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_390_0,
        encoder_layers_25_mlp_fc1_weight,

        encoder_layers_25_mlp_fc1_bias,

        gemm_rcr_bias_gelu_391_0,
        global_workspace_,
        1,

        &reshape_389_0_dim_0,

        &reshape_389_0_dim_1,

        &reshape_389_0_dim_2,


        &encoder_layers_25_mlp_fc1_weight_dim_0,

        &encoder_layers_25_mlp_fc1_weight_dim_1,


        &reshape_389_0_dim_0,

        &reshape_389_0_dim_1,

        &encoder_layers_25_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_391_0,
        encoder_layers_25_mlp_fc2_weight,
        encoder_layers_25_mlp_fc2_bias,
        reshape_389_0,

        gemm_rcr_bias_add_392_0,
        global_workspace_,

     1,


        &reshape_389_0_dim_0,

        &reshape_389_0_dim_1,

        &encoder_layers_25_mlp_fc1_weight_dim_0,


        &encoder_layers_25_mlp_fc2_weight_dim_0,

        &encoder_layers_25_mlp_fc2_weight_dim_1,


        &reshape_389_0_dim_0,

        &reshape_389_0_dim_1,

        &encoder_layers_25_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_389_0_dim_0;

        M *= reshape_389_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_25_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_394_0, gemm_rcr_bias_add_392_0, encoder_layers_26_ln1_weight, encoder_layers_26_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_permute_5(

        reshape_394_0,
        encoder_layers_26_mha_qkv_weight,

        encoder_layers_26_mha_qkv_bias,

        reshape_396_0,
        global_workspace_,
        1,

        &reshape_394_0_dim_0,

        &reshape_394_0_dim_1,


        &encoder_layers_26_mha_qkv_weight_dim_0,

        &encoder_layers_26_mha_qkv_weight_dim_1,


        &reshape_394_0_dim_0,

        &encoder_layers_26_mha_qkv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_397_1,
      split_397_2,
      split_397_0
    };


      int64_t *split_397_1_shape[] = {
        &split_397_1_dim_0, &reshape_396_0_dim_1, &reshape_396_0_dim_2, &reshape_396_0_dim_3, &reshape_396_0_dim_4
      };
      int64_t *split_397_2_shape[] = {
        &split_397_2_dim_0, &reshape_396_0_dim_1, &reshape_396_0_dim_2, &reshape_396_0_dim_3, &reshape_396_0_dim_4
      };
      int64_t *split_397_0_shape[] = {
        &split_397_0_dim_0, &reshape_396_0_dim_1, &reshape_396_0_dim_2, &reshape_396_0_dim_3, &reshape_396_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_397_1_shape, split_397_2_shape, split_397_0_shape
    };

    const int64_t reshape_396_0_shape[] = {
      reshape_396_0_dim_0, reshape_396_0_dim_1, reshape_396_0_dim_2, reshape_396_0_dim_3, reshape_396_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_396_0,
        reshape_396_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_397_0_dim_0,
        &reshape_396_0_dim_1,
        &reshape_396_0_dim_2,
        &reshape_396_0_dim_3,
        &reshape_396_0_dim_4,
        &reshape_398_0_dim_0,
        &reshape_398_0_dim_1,
        &reshape_398_0_dim_2,
        &reshape_398_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_397_1_dim_0,
        &reshape_396_0_dim_1,
        &reshape_396_0_dim_2,
        &reshape_396_0_dim_3,
        &reshape_396_0_dim_4,
        &reshape_399_0_dim_0,
        &reshape_399_0_dim_1,
        &reshape_399_0_dim_2,
        &reshape_399_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_8(
        &split_397_2_dim_0,
        &reshape_396_0_dim_1,
        &reshape_396_0_dim_2,
        &reshape_396_0_dim_3,
        &reshape_396_0_dim_4,
        &reshape_400_0_dim_0,
        &reshape_400_0_dim_1,
        &reshape_400_0_dim_2,
        &reshape_400_0_dim_3
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_11(
        mem_eff_attention_401_0,
        reshape_399_0, reshape_400_0, reshape_398_0,
        &reshape_399_0_dim_0,
        &reshape_398_0_dim_2,
        &reshape_399_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_401_0,
        encoder_layers_26_mha_proj_weight,
        encoder_layers_26_mha_proj_bias,
        gemm_rcr_bias_add_392_0,

        reshape_404_0,
        global_workspace_,

     1,


        &reshape_402_0_dim_0,

        &reshape_402_0_dim_1,


        &encoder_layers_26_mha_proj_weight_dim_0,

        &encoder_layers_26_mha_proj_weight_dim_1,


        &reshape_402_0_dim_0,

        &encoder_layers_26_mha_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_404_0_dim_0;

        M *= reshape_404_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_404_0_dim_2;

    
        layernorm_15(
           layernorm_405_0, reshape_404_0, encoder_layers_26_ln2_weight, encoder_layers_26_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_405_0,
        encoder_layers_26_mlp_fc1_weight,

        encoder_layers_26_mlp_fc1_bias,

        gemm_rcr_bias_gelu_406_0,
        global_workspace_,
        1,

        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &reshape_404_0_dim_2,


        &encoder_layers_26_mlp_fc1_weight_dim_0,

        &encoder_layers_26_mlp_fc1_weight_dim_1,


        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &encoder_layers_26_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_406_0,
        encoder_layers_26_mlp_fc2_weight,
        encoder_layers_26_mlp_fc2_bias,
        reshape_404_0,

        gemm_rcr_bias_add_407_0,
        global_workspace_,

     1,


        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &encoder_layers_26_mlp_fc1_weight_dim_0,


        &encoder_layers_26_mlp_fc2_weight_dim_0,

        &encoder_layers_26_mlp_fc2_weight_dim_1,


        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &encoder_layers_26_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_404_0_dim_0;

        M *= reshape_404_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_26_mlp_fc2_weight_dim_0;

    
        layernorm_15(
           layernorm_408_0, gemm_rcr_bias_add_407_0, encoder_ln_weight, encoder_ln_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_409(

        layernorm_408_0,
        pool_kv_weight,

        pool_kv_bias,

        reshape_410_0,
        global_workspace_,
        1,

        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &encoder_layers_26_mlp_fc2_weight_dim_0,


        &pool_kv_weight_dim_0,

        &pool_kv_weight_dim_1,


        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &pool_kv_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
{
    int dims[] = {2,0,3,1,4};
    permute_411(
        permute_411_0,
        reshape_410_0,

        &reshape_410_0_dim_0,

        &reshape_410_0_dim_1,

        &reshape_410_0_dim_2,

        &reshape_410_0_dim_3,

        &reshape_410_0_dim_4,

        dims,
        stream
    );
}
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
  {

    void *outputs[] = {
      split_412_1,
      split_412_0
    };


      int64_t *split_412_1_shape[] = {
        &split_412_1_dim_0, &reshape_410_0_dim_0, &reshape_410_0_dim_3, &reshape_410_0_dim_1, &reshape_410_0_dim_4
      };
      int64_t *split_412_0_shape[] = {
        &split_412_0_dim_0, &reshape_410_0_dim_0, &reshape_410_0_dim_3, &reshape_410_0_dim_1, &reshape_410_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_412_1_shape, split_412_0_shape
    };

    const int64_t permute_411_0_shape[] = {
      reshape_410_0_dim_2, reshape_410_0_dim_0, reshape_410_0_dim_3, reshape_410_0_dim_1, reshape_410_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1
    };

    bool output_masks[] = {
      true, true
    };

    split_412(
        outputs,
        output_shapes,
        output_masks,
        permute_411_0,
        permute_411_0_shape,
        2/*real_num_splits*/,
        2/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_squeeze_413(
        &split_412_0_dim_0,
        &reshape_410_0_dim_0,
        &reshape_410_0_dim_3,
        &reshape_410_0_dim_1,
        &reshape_410_0_dim_4,
        &reshape_410_0_dim_0,
        &reshape_410_0_dim_3,
        &reshape_410_0_dim_1,
        &reshape_410_0_dim_4
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_squeeze_413(
        &split_412_1_dim_0,
        &reshape_410_0_dim_0,
        &reshape_410_0_dim_3,
        &reshape_410_0_dim_1,
        &reshape_410_0_dim_4,
        &reshape_410_0_dim_0,
        &reshape_410_0_dim_3,
        &reshape_410_0_dim_1,
        &reshape_410_0_dim_4
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    mem_eff_attention_418(
        mem_eff_attention_418_0,
        reshape_417_0, squeeze_414_0, squeeze_413_0,
        &reshape_417_0_dim_0,
        &reshape_410_0_dim_1,
        &reshape_417_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    ait_reshape_426(
        &reshape_417_0_dim_0,
        &reshape_417_0_dim_2,
        &reshape_417_0_dim_1,
        &reshape_410_0_dim_4,
        &reshape_417_0_dim_0,
        &reshape_417_0_dim_1,
        &reshape_417_0_dim_2,
        &reshape_410_0_dim_4
    );
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_421(

        reshape_426_0,
        pool_proj_weight,

        pool_proj_bias,

        gemm_rcr_bias_421_0,
        global_workspace_,
        1,

        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &reshape_420_0_dim_2,


        &pool_proj_weight_dim_0,

        &pool_proj_weight_dim_1,


        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &pool_proj_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
      {
        
        int64_t M = 1;

        M *= reshape_420_0_dim_0;

        M *= reshape_420_0_dim_1;

    

        int64_t N = 1;

        N *= pool_proj_weight_dim_0;

    
        layernorm_422(
           layernorm_422_0, gemm_rcr_bias_421_0, pool_ln_weight, pool_ln_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_gelu_423(

        layernorm_422_0,
        pool_mlp_fc1_weight,

        pool_mlp_fc1_bias,

        gemm_rcr_bias_gelu_423_0,
        global_workspace_,
        1,

        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &pool_proj_weight_dim_0,


        &pool_mlp_fc1_weight_dim_0,

        &pool_mlp_fc1_weight_dim_1,


        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &pool_mlp_fc1_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
  
    {
    

    gemm_rcr_bias_add_424(

        gemm_rcr_bias_gelu_423_0,
        pool_mlp_fc2_weight,
        pool_mlp_fc2_bias,
        gemm_rcr_bias_421_0,

        output_0,
        global_workspace_,

     2,


        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &pool_mlp_fc1_weight_dim_0,


        &pool_mlp_fc2_weight_dim_0,

        &pool_mlp_fc2_weight_dim_1,


        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &pool_mlp_fc2_weight_dim_0,

        stream
    );
    }
      DeviceCheckLastError(__FILE__, __LINE__);
  
    }




    void ProfileImpl(StreamType stream, size_t iters, const std::string& filename) {
#ifdef OPTIMIZE_FOR_COMPILATION_TIME
      throw std::runtime_error("Profile is disabled, please recompile without OPTIMIZE_FOR_COMPILE_TIME flag");
#else
      std::ofstream ss(filename);
      if (!ss) {
        throw std::runtime_error(std::string("Could not open file ") + filename);
      }

      int deviceId;
      char* L2CacheSlab = nullptr;
      DevicePropertyType deviceProperties;
      GetDevice(&deviceId);
      GetDeviceProperties(&deviceProperties, deviceId);
      const size_t L2SizeInBytes = deviceProperties.l2CacheSize;
      DeviceMalloc((void**) &L2CacheSlab, L2SizeInBytes);

      ss << "{\n";
      
      {
        std::cout << "Profiling: " << "conv2d_bias_few_channels_0" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    conv2d_bias_few_channels_0(
        input0,
        patch_embed_proj_weight,
        conv2d_bias_few_channels_0_0,

        patch_embed_proj_bias,

        global_workspace_,
        &input0_dim_0,
        &patch_embed_proj_weight_dim_0,
        &input0_dim_3,
        &patch_embed_proj_weight_dim_1,
        &patch_embed_proj_weight_dim_2,
        &input0_dim_1,
        &input0_dim_2,
        &input0_dim_0,
        &conv2d_bias_few_channels_0_0_dim_1,
        &conv2d_bias_few_channels_0_0_dim_2,
        14,
        1,
        0,
        14,
        1,
        0,
        stream
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "conv2d_bias_few_channels_0" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"384\", \"384\", \"3\"], [\"1152\", \"14\", \"14\", \"3\"], [\"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"27\", \"27\", \"1152\"]]"
        
          << ", \"auto_padding\": " << "\"False\""
        
          << ", \"dilate\": " << "\"1\""
        
          << ", \"pad\": " << "\"0\""
        
          << ", \"stride\": " << "\"14\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "fused_elementwise_425" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
        int64_t fused_elementwise_425_n_elements = 8 * 729 * 1152;
        invoke_fused_elementwise_425(elementwise_2_0, conv2d_bias_few_channels_0_0,pos_emb_pos_emb,   fused_elementwise_425_n_elements, stream);
    }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "fused_elementwise_425" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"1\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"func\": " << "\"[<FuncEnum.ADD: 1>]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_3" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_1_0_dim_0;

        M *= reshape_1_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_1_0_dim_2;

    
        layernorm_3(
           reshape_4_0, elementwise_2_0, encoder_layers_0_ln1_weight, encoder_layers_0_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_3" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_5" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_4_0,
        encoder_layers_0_mha_qkv_weight,

        encoder_layers_0_mha_qkv_bias,

        reshape_6_0,
        global_workspace_,
        1,

        &reshape_4_0_dim_0,

        &reshape_4_0_dim_1,


        &encoder_layers_0_mha_qkv_weight_dim_0,

        &encoder_layers_0_mha_qkv_weight_dim_1,


        &reshape_4_0_dim_0,

        &encoder_layers_0_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_5" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_7" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_7_0,
      split_7_1,
      split_7_2
    };


      int64_t *split_7_0_shape[] = {
        &split_7_0_dim_0, &reshape_6_0_dim_1, &reshape_6_0_dim_2, &reshape_6_0_dim_3, &reshape_6_0_dim_4
      };
      int64_t *split_7_1_shape[] = {
        &split_7_1_dim_0, &reshape_6_0_dim_1, &reshape_6_0_dim_2, &reshape_6_0_dim_3, &reshape_6_0_dim_4
      };
      int64_t *split_7_2_shape[] = {
        &split_7_2_dim_0, &reshape_6_0_dim_1, &reshape_6_0_dim_2, &reshape_6_0_dim_3, &reshape_6_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_7_0_shape, split_7_1_shape, split_7_2_shape
    };

    const int64_t reshape_6_0_shape[] = {
      reshape_6_0_dim_0, reshape_6_0_dim_1, reshape_6_0_dim_2, reshape_6_0_dim_3, reshape_6_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_6_0,
        reshape_6_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_7" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_8" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_7_0_dim_0,
        &reshape_6_0_dim_1,
        &reshape_6_0_dim_2,
        &reshape_6_0_dim_3,
        &reshape_6_0_dim_4,
        &reshape_8_0_dim_0,
        &reshape_8_0_dim_1,
        &reshape_8_0_dim_2,
        &reshape_8_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_8" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_9" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_7_1_dim_0,
        &reshape_6_0_dim_1,
        &reshape_6_0_dim_2,
        &reshape_6_0_dim_3,
        &reshape_6_0_dim_4,
        &reshape_9_0_dim_0,
        &reshape_9_0_dim_1,
        &reshape_9_0_dim_2,
        &reshape_9_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_9" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_10" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_7_2_dim_0,
        &reshape_6_0_dim_1,
        &reshape_6_0_dim_2,
        &reshape_6_0_dim_3,
        &reshape_6_0_dim_4,
        &reshape_10_0_dim_0,
        &reshape_10_0_dim_1,
        &reshape_10_0_dim_2,
        &reshape_10_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_10" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_11" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_11_0,
        reshape_8_0, reshape_9_0, reshape_10_0,
        &reshape_8_0_dim_0,
        &reshape_10_0_dim_2,
        &reshape_8_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_11" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_13" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_11_0,
        encoder_layers_0_mha_proj_weight,
        encoder_layers_0_mha_proj_bias,
        elementwise_2_0,

        reshape_14_0,
        global_workspace_,

     1,


        &reshape_12_0_dim_0,

        &reshape_12_0_dim_1,


        &encoder_layers_0_mha_proj_weight_dim_0,

        &encoder_layers_0_mha_proj_weight_dim_1,


        &reshape_12_0_dim_0,

        &encoder_layers_0_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_13" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_15" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_14_0_dim_0;

        M *= reshape_14_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_14_0_dim_2;

    
        layernorm_15(
           layernorm_15_0, reshape_14_0, encoder_layers_0_ln2_weight, encoder_layers_0_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_15" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_16" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_15_0,
        encoder_layers_0_mlp_fc1_weight,

        encoder_layers_0_mlp_fc1_bias,

        gemm_rcr_bias_gelu_16_0,
        global_workspace_,
        1,

        &reshape_14_0_dim_0,

        &reshape_14_0_dim_1,

        &reshape_14_0_dim_2,


        &encoder_layers_0_mlp_fc1_weight_dim_0,

        &encoder_layers_0_mlp_fc1_weight_dim_1,


        &reshape_14_0_dim_0,

        &reshape_14_0_dim_1,

        &encoder_layers_0_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_16" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_17" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_16_0,
        encoder_layers_0_mlp_fc2_weight,
        encoder_layers_0_mlp_fc2_bias,
        reshape_14_0,

        gemm_rcr_bias_add_17_0,
        global_workspace_,

     1,


        &reshape_14_0_dim_0,

        &reshape_14_0_dim_1,

        &encoder_layers_0_mlp_fc1_weight_dim_0,


        &encoder_layers_0_mlp_fc2_weight_dim_0,

        &encoder_layers_0_mlp_fc2_weight_dim_1,


        &reshape_14_0_dim_0,

        &reshape_14_0_dim_1,

        &encoder_layers_0_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_17" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_18" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_14_0_dim_0;

        M *= reshape_14_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_0_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_19_0, gemm_rcr_bias_add_17_0, encoder_layers_1_ln1_weight, encoder_layers_1_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_18" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_20" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_19_0,
        encoder_layers_1_mha_qkv_weight,

        encoder_layers_1_mha_qkv_bias,

        reshape_21_0,
        global_workspace_,
        1,

        &reshape_19_0_dim_0,

        &reshape_19_0_dim_1,


        &encoder_layers_1_mha_qkv_weight_dim_0,

        &encoder_layers_1_mha_qkv_weight_dim_1,


        &reshape_19_0_dim_0,

        &encoder_layers_1_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_20" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_22" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_22_0,
      split_22_2,
      split_22_1
    };


      int64_t *split_22_0_shape[] = {
        &split_22_0_dim_0, &reshape_21_0_dim_1, &reshape_21_0_dim_2, &reshape_21_0_dim_3, &reshape_21_0_dim_4
      };
      int64_t *split_22_2_shape[] = {
        &split_22_2_dim_0, &reshape_21_0_dim_1, &reshape_21_0_dim_2, &reshape_21_0_dim_3, &reshape_21_0_dim_4
      };
      int64_t *split_22_1_shape[] = {
        &split_22_1_dim_0, &reshape_21_0_dim_1, &reshape_21_0_dim_2, &reshape_21_0_dim_3, &reshape_21_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_22_0_shape, split_22_2_shape, split_22_1_shape
    };

    const int64_t reshape_21_0_shape[] = {
      reshape_21_0_dim_0, reshape_21_0_dim_1, reshape_21_0_dim_2, reshape_21_0_dim_3, reshape_21_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_21_0,
        reshape_21_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_22" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_23" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_22_0_dim_0,
        &reshape_21_0_dim_1,
        &reshape_21_0_dim_2,
        &reshape_21_0_dim_3,
        &reshape_21_0_dim_4,
        &reshape_23_0_dim_0,
        &reshape_23_0_dim_1,
        &reshape_23_0_dim_2,
        &reshape_23_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_23" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_24" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_22_1_dim_0,
        &reshape_21_0_dim_1,
        &reshape_21_0_dim_2,
        &reshape_21_0_dim_3,
        &reshape_21_0_dim_4,
        &reshape_24_0_dim_0,
        &reshape_24_0_dim_1,
        &reshape_24_0_dim_2,
        &reshape_24_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_24" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_25" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_22_2_dim_0,
        &reshape_21_0_dim_1,
        &reshape_21_0_dim_2,
        &reshape_21_0_dim_3,
        &reshape_21_0_dim_4,
        &reshape_25_0_dim_0,
        &reshape_25_0_dim_1,
        &reshape_25_0_dim_2,
        &reshape_25_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_25" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_26" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_26_0,
        reshape_23_0, reshape_25_0, reshape_24_0,
        &reshape_23_0_dim_0,
        &reshape_24_0_dim_2,
        &reshape_23_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_26" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_28" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_26_0,
        encoder_layers_1_mha_proj_weight,
        encoder_layers_1_mha_proj_bias,
        gemm_rcr_bias_add_17_0,

        reshape_29_0,
        global_workspace_,

     1,


        &reshape_27_0_dim_0,

        &reshape_27_0_dim_1,


        &encoder_layers_1_mha_proj_weight_dim_0,

        &encoder_layers_1_mha_proj_weight_dim_1,


        &reshape_27_0_dim_0,

        &encoder_layers_1_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_28" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_30" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_29_0_dim_0;

        M *= reshape_29_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_29_0_dim_2;

    
        layernorm_15(
           layernorm_30_0, reshape_29_0, encoder_layers_1_ln2_weight, encoder_layers_1_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_30" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_31" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_30_0,
        encoder_layers_1_mlp_fc1_weight,

        encoder_layers_1_mlp_fc1_bias,

        gemm_rcr_bias_gelu_31_0,
        global_workspace_,
        1,

        &reshape_29_0_dim_0,

        &reshape_29_0_dim_1,

        &reshape_29_0_dim_2,


        &encoder_layers_1_mlp_fc1_weight_dim_0,

        &encoder_layers_1_mlp_fc1_weight_dim_1,


        &reshape_29_0_dim_0,

        &reshape_29_0_dim_1,

        &encoder_layers_1_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_31" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_32" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_31_0,
        encoder_layers_1_mlp_fc2_weight,
        encoder_layers_1_mlp_fc2_bias,
        reshape_29_0,

        gemm_rcr_bias_add_32_0,
        global_workspace_,

     1,


        &reshape_29_0_dim_0,

        &reshape_29_0_dim_1,

        &encoder_layers_1_mlp_fc1_weight_dim_0,


        &encoder_layers_1_mlp_fc2_weight_dim_0,

        &encoder_layers_1_mlp_fc2_weight_dim_1,


        &reshape_29_0_dim_0,

        &reshape_29_0_dim_1,

        &encoder_layers_1_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_32" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_33" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_29_0_dim_0;

        M *= reshape_29_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_1_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_34_0, gemm_rcr_bias_add_32_0, encoder_layers_2_ln1_weight, encoder_layers_2_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_33" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_35" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_34_0,
        encoder_layers_2_mha_qkv_weight,

        encoder_layers_2_mha_qkv_bias,

        reshape_36_0,
        global_workspace_,
        1,

        &reshape_34_0_dim_0,

        &reshape_34_0_dim_1,


        &encoder_layers_2_mha_qkv_weight_dim_0,

        &encoder_layers_2_mha_qkv_weight_dim_1,


        &reshape_34_0_dim_0,

        &encoder_layers_2_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_35" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_37" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_37_1,
      split_37_0,
      split_37_2
    };


      int64_t *split_37_1_shape[] = {
        &split_37_1_dim_0, &reshape_36_0_dim_1, &reshape_36_0_dim_2, &reshape_36_0_dim_3, &reshape_36_0_dim_4
      };
      int64_t *split_37_0_shape[] = {
        &split_37_0_dim_0, &reshape_36_0_dim_1, &reshape_36_0_dim_2, &reshape_36_0_dim_3, &reshape_36_0_dim_4
      };
      int64_t *split_37_2_shape[] = {
        &split_37_2_dim_0, &reshape_36_0_dim_1, &reshape_36_0_dim_2, &reshape_36_0_dim_3, &reshape_36_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_37_1_shape, split_37_0_shape, split_37_2_shape
    };

    const int64_t reshape_36_0_shape[] = {
      reshape_36_0_dim_0, reshape_36_0_dim_1, reshape_36_0_dim_2, reshape_36_0_dim_3, reshape_36_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_36_0,
        reshape_36_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_37" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_38" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_37_0_dim_0,
        &reshape_36_0_dim_1,
        &reshape_36_0_dim_2,
        &reshape_36_0_dim_3,
        &reshape_36_0_dim_4,
        &reshape_38_0_dim_0,
        &reshape_38_0_dim_1,
        &reshape_38_0_dim_2,
        &reshape_38_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_38" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_39" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_37_1_dim_0,
        &reshape_36_0_dim_1,
        &reshape_36_0_dim_2,
        &reshape_36_0_dim_3,
        &reshape_36_0_dim_4,
        &reshape_39_0_dim_0,
        &reshape_39_0_dim_1,
        &reshape_39_0_dim_2,
        &reshape_39_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_39" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_40" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_37_2_dim_0,
        &reshape_36_0_dim_1,
        &reshape_36_0_dim_2,
        &reshape_36_0_dim_3,
        &reshape_36_0_dim_4,
        &reshape_40_0_dim_0,
        &reshape_40_0_dim_1,
        &reshape_40_0_dim_2,
        &reshape_40_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_40" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_41" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_41_0,
        reshape_39_0, reshape_38_0, reshape_40_0,
        &reshape_39_0_dim_0,
        &reshape_40_0_dim_2,
        &reshape_39_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_41" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_43" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_41_0,
        encoder_layers_2_mha_proj_weight,
        encoder_layers_2_mha_proj_bias,
        gemm_rcr_bias_add_32_0,

        reshape_44_0,
        global_workspace_,

     1,


        &reshape_42_0_dim_0,

        &reshape_42_0_dim_1,


        &encoder_layers_2_mha_proj_weight_dim_0,

        &encoder_layers_2_mha_proj_weight_dim_1,


        &reshape_42_0_dim_0,

        &encoder_layers_2_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_43" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_45" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_44_0_dim_0;

        M *= reshape_44_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_44_0_dim_2;

    
        layernorm_15(
           layernorm_45_0, reshape_44_0, encoder_layers_2_ln2_weight, encoder_layers_2_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_45" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_46" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_45_0,
        encoder_layers_2_mlp_fc1_weight,

        encoder_layers_2_mlp_fc1_bias,

        gemm_rcr_bias_gelu_46_0,
        global_workspace_,
        1,

        &reshape_44_0_dim_0,

        &reshape_44_0_dim_1,

        &reshape_44_0_dim_2,


        &encoder_layers_2_mlp_fc1_weight_dim_0,

        &encoder_layers_2_mlp_fc1_weight_dim_1,


        &reshape_44_0_dim_0,

        &reshape_44_0_dim_1,

        &encoder_layers_2_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_46" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_47" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_46_0,
        encoder_layers_2_mlp_fc2_weight,
        encoder_layers_2_mlp_fc2_bias,
        reshape_44_0,

        gemm_rcr_bias_add_47_0,
        global_workspace_,

     1,


        &reshape_44_0_dim_0,

        &reshape_44_0_dim_1,

        &encoder_layers_2_mlp_fc1_weight_dim_0,


        &encoder_layers_2_mlp_fc2_weight_dim_0,

        &encoder_layers_2_mlp_fc2_weight_dim_1,


        &reshape_44_0_dim_0,

        &reshape_44_0_dim_1,

        &encoder_layers_2_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_47" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_48" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_44_0_dim_0;

        M *= reshape_44_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_2_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_49_0, gemm_rcr_bias_add_47_0, encoder_layers_3_ln1_weight, encoder_layers_3_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_48" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_50" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_49_0,
        encoder_layers_3_mha_qkv_weight,

        encoder_layers_3_mha_qkv_bias,

        reshape_51_0,
        global_workspace_,
        1,

        &reshape_49_0_dim_0,

        &reshape_49_0_dim_1,


        &encoder_layers_3_mha_qkv_weight_dim_0,

        &encoder_layers_3_mha_qkv_weight_dim_1,


        &reshape_49_0_dim_0,

        &encoder_layers_3_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_50" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_52" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_52_2,
      split_52_1,
      split_52_0
    };


      int64_t *split_52_2_shape[] = {
        &split_52_2_dim_0, &reshape_51_0_dim_1, &reshape_51_0_dim_2, &reshape_51_0_dim_3, &reshape_51_0_dim_4
      };
      int64_t *split_52_1_shape[] = {
        &split_52_1_dim_0, &reshape_51_0_dim_1, &reshape_51_0_dim_2, &reshape_51_0_dim_3, &reshape_51_0_dim_4
      };
      int64_t *split_52_0_shape[] = {
        &split_52_0_dim_0, &reshape_51_0_dim_1, &reshape_51_0_dim_2, &reshape_51_0_dim_3, &reshape_51_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_52_2_shape, split_52_1_shape, split_52_0_shape
    };

    const int64_t reshape_51_0_shape[] = {
      reshape_51_0_dim_0, reshape_51_0_dim_1, reshape_51_0_dim_2, reshape_51_0_dim_3, reshape_51_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_51_0,
        reshape_51_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_52" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_53" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_52_0_dim_0,
        &reshape_51_0_dim_1,
        &reshape_51_0_dim_2,
        &reshape_51_0_dim_3,
        &reshape_51_0_dim_4,
        &reshape_53_0_dim_0,
        &reshape_53_0_dim_1,
        &reshape_53_0_dim_2,
        &reshape_53_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_53" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_54" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_52_1_dim_0,
        &reshape_51_0_dim_1,
        &reshape_51_0_dim_2,
        &reshape_51_0_dim_3,
        &reshape_51_0_dim_4,
        &reshape_54_0_dim_0,
        &reshape_54_0_dim_1,
        &reshape_54_0_dim_2,
        &reshape_54_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_54" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_55" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_52_2_dim_0,
        &reshape_51_0_dim_1,
        &reshape_51_0_dim_2,
        &reshape_51_0_dim_3,
        &reshape_51_0_dim_4,
        &reshape_55_0_dim_0,
        &reshape_55_0_dim_1,
        &reshape_55_0_dim_2,
        &reshape_55_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_55" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_56" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_56_0,
        reshape_55_0, reshape_54_0, reshape_53_0,
        &reshape_55_0_dim_0,
        &reshape_53_0_dim_2,
        &reshape_55_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_56" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_58" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_56_0,
        encoder_layers_3_mha_proj_weight,
        encoder_layers_3_mha_proj_bias,
        gemm_rcr_bias_add_47_0,

        reshape_59_0,
        global_workspace_,

     1,


        &reshape_57_0_dim_0,

        &reshape_57_0_dim_1,


        &encoder_layers_3_mha_proj_weight_dim_0,

        &encoder_layers_3_mha_proj_weight_dim_1,


        &reshape_57_0_dim_0,

        &encoder_layers_3_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_58" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_60" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_59_0_dim_0;

        M *= reshape_59_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_59_0_dim_2;

    
        layernorm_15(
           layernorm_60_0, reshape_59_0, encoder_layers_3_ln2_weight, encoder_layers_3_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_60" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_61" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_60_0,
        encoder_layers_3_mlp_fc1_weight,

        encoder_layers_3_mlp_fc1_bias,

        gemm_rcr_bias_gelu_61_0,
        global_workspace_,
        1,

        &reshape_59_0_dim_0,

        &reshape_59_0_dim_1,

        &reshape_59_0_dim_2,


        &encoder_layers_3_mlp_fc1_weight_dim_0,

        &encoder_layers_3_mlp_fc1_weight_dim_1,


        &reshape_59_0_dim_0,

        &reshape_59_0_dim_1,

        &encoder_layers_3_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_61" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_62" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_61_0,
        encoder_layers_3_mlp_fc2_weight,
        encoder_layers_3_mlp_fc2_bias,
        reshape_59_0,

        gemm_rcr_bias_add_62_0,
        global_workspace_,

     1,


        &reshape_59_0_dim_0,

        &reshape_59_0_dim_1,

        &encoder_layers_3_mlp_fc1_weight_dim_0,


        &encoder_layers_3_mlp_fc2_weight_dim_0,

        &encoder_layers_3_mlp_fc2_weight_dim_1,


        &reshape_59_0_dim_0,

        &reshape_59_0_dim_1,

        &encoder_layers_3_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_62" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_63" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_59_0_dim_0;

        M *= reshape_59_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_3_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_64_0, gemm_rcr_bias_add_62_0, encoder_layers_4_ln1_weight, encoder_layers_4_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_63" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_65" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_64_0,
        encoder_layers_4_mha_qkv_weight,

        encoder_layers_4_mha_qkv_bias,

        reshape_66_0,
        global_workspace_,
        1,

        &reshape_64_0_dim_0,

        &reshape_64_0_dim_1,


        &encoder_layers_4_mha_qkv_weight_dim_0,

        &encoder_layers_4_mha_qkv_weight_dim_1,


        &reshape_64_0_dim_0,

        &encoder_layers_4_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_65" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_67" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_67_0,
      split_67_1,
      split_67_2
    };


      int64_t *split_67_0_shape[] = {
        &split_67_0_dim_0, &reshape_66_0_dim_1, &reshape_66_0_dim_2, &reshape_66_0_dim_3, &reshape_66_0_dim_4
      };
      int64_t *split_67_1_shape[] = {
        &split_67_1_dim_0, &reshape_66_0_dim_1, &reshape_66_0_dim_2, &reshape_66_0_dim_3, &reshape_66_0_dim_4
      };
      int64_t *split_67_2_shape[] = {
        &split_67_2_dim_0, &reshape_66_0_dim_1, &reshape_66_0_dim_2, &reshape_66_0_dim_3, &reshape_66_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_67_0_shape, split_67_1_shape, split_67_2_shape
    };

    const int64_t reshape_66_0_shape[] = {
      reshape_66_0_dim_0, reshape_66_0_dim_1, reshape_66_0_dim_2, reshape_66_0_dim_3, reshape_66_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_66_0,
        reshape_66_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_67" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_68" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_67_0_dim_0,
        &reshape_66_0_dim_1,
        &reshape_66_0_dim_2,
        &reshape_66_0_dim_3,
        &reshape_66_0_dim_4,
        &reshape_68_0_dim_0,
        &reshape_68_0_dim_1,
        &reshape_68_0_dim_2,
        &reshape_68_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_68" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_69" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_67_1_dim_0,
        &reshape_66_0_dim_1,
        &reshape_66_0_dim_2,
        &reshape_66_0_dim_3,
        &reshape_66_0_dim_4,
        &reshape_69_0_dim_0,
        &reshape_69_0_dim_1,
        &reshape_69_0_dim_2,
        &reshape_69_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_69" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_70" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_67_2_dim_0,
        &reshape_66_0_dim_1,
        &reshape_66_0_dim_2,
        &reshape_66_0_dim_3,
        &reshape_66_0_dim_4,
        &reshape_70_0_dim_0,
        &reshape_70_0_dim_1,
        &reshape_70_0_dim_2,
        &reshape_70_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_70" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_71" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_71_0,
        reshape_68_0, reshape_69_0, reshape_70_0,
        &reshape_68_0_dim_0,
        &reshape_70_0_dim_2,
        &reshape_68_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_71" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_73" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_71_0,
        encoder_layers_4_mha_proj_weight,
        encoder_layers_4_mha_proj_bias,
        gemm_rcr_bias_add_62_0,

        reshape_74_0,
        global_workspace_,

     1,


        &reshape_72_0_dim_0,

        &reshape_72_0_dim_1,


        &encoder_layers_4_mha_proj_weight_dim_0,

        &encoder_layers_4_mha_proj_weight_dim_1,


        &reshape_72_0_dim_0,

        &encoder_layers_4_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_73" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_75" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_74_0_dim_0;

        M *= reshape_74_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_74_0_dim_2;

    
        layernorm_15(
           layernorm_75_0, reshape_74_0, encoder_layers_4_ln2_weight, encoder_layers_4_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_75" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_76" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_75_0,
        encoder_layers_4_mlp_fc1_weight,

        encoder_layers_4_mlp_fc1_bias,

        gemm_rcr_bias_gelu_76_0,
        global_workspace_,
        1,

        &reshape_74_0_dim_0,

        &reshape_74_0_dim_1,

        &reshape_74_0_dim_2,


        &encoder_layers_4_mlp_fc1_weight_dim_0,

        &encoder_layers_4_mlp_fc1_weight_dim_1,


        &reshape_74_0_dim_0,

        &reshape_74_0_dim_1,

        &encoder_layers_4_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_76" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_77" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_76_0,
        encoder_layers_4_mlp_fc2_weight,
        encoder_layers_4_mlp_fc2_bias,
        reshape_74_0,

        gemm_rcr_bias_add_77_0,
        global_workspace_,

     1,


        &reshape_74_0_dim_0,

        &reshape_74_0_dim_1,

        &encoder_layers_4_mlp_fc1_weight_dim_0,


        &encoder_layers_4_mlp_fc2_weight_dim_0,

        &encoder_layers_4_mlp_fc2_weight_dim_1,


        &reshape_74_0_dim_0,

        &reshape_74_0_dim_1,

        &encoder_layers_4_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_77" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_78" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_74_0_dim_0;

        M *= reshape_74_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_4_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_79_0, gemm_rcr_bias_add_77_0, encoder_layers_5_ln1_weight, encoder_layers_5_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_78" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_80" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_79_0,
        encoder_layers_5_mha_qkv_weight,

        encoder_layers_5_mha_qkv_bias,

        reshape_81_0,
        global_workspace_,
        1,

        &reshape_79_0_dim_0,

        &reshape_79_0_dim_1,


        &encoder_layers_5_mha_qkv_weight_dim_0,

        &encoder_layers_5_mha_qkv_weight_dim_1,


        &reshape_79_0_dim_0,

        &encoder_layers_5_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_80" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_82" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_82_0,
      split_82_1,
      split_82_2
    };


      int64_t *split_82_0_shape[] = {
        &split_82_0_dim_0, &reshape_81_0_dim_1, &reshape_81_0_dim_2, &reshape_81_0_dim_3, &reshape_81_0_dim_4
      };
      int64_t *split_82_1_shape[] = {
        &split_82_1_dim_0, &reshape_81_0_dim_1, &reshape_81_0_dim_2, &reshape_81_0_dim_3, &reshape_81_0_dim_4
      };
      int64_t *split_82_2_shape[] = {
        &split_82_2_dim_0, &reshape_81_0_dim_1, &reshape_81_0_dim_2, &reshape_81_0_dim_3, &reshape_81_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_82_0_shape, split_82_1_shape, split_82_2_shape
    };

    const int64_t reshape_81_0_shape[] = {
      reshape_81_0_dim_0, reshape_81_0_dim_1, reshape_81_0_dim_2, reshape_81_0_dim_3, reshape_81_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_81_0,
        reshape_81_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_82" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_83" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_82_0_dim_0,
        &reshape_81_0_dim_1,
        &reshape_81_0_dim_2,
        &reshape_81_0_dim_3,
        &reshape_81_0_dim_4,
        &reshape_83_0_dim_0,
        &reshape_83_0_dim_1,
        &reshape_83_0_dim_2,
        &reshape_83_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_83" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_84" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_82_1_dim_0,
        &reshape_81_0_dim_1,
        &reshape_81_0_dim_2,
        &reshape_81_0_dim_3,
        &reshape_81_0_dim_4,
        &reshape_84_0_dim_0,
        &reshape_84_0_dim_1,
        &reshape_84_0_dim_2,
        &reshape_84_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_84" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_85" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_82_2_dim_0,
        &reshape_81_0_dim_1,
        &reshape_81_0_dim_2,
        &reshape_81_0_dim_3,
        &reshape_81_0_dim_4,
        &reshape_85_0_dim_0,
        &reshape_85_0_dim_1,
        &reshape_85_0_dim_2,
        &reshape_85_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_85" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_86" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_86_0,
        reshape_83_0, reshape_84_0, reshape_85_0,
        &reshape_83_0_dim_0,
        &reshape_85_0_dim_2,
        &reshape_83_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_86" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_88" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_86_0,
        encoder_layers_5_mha_proj_weight,
        encoder_layers_5_mha_proj_bias,
        gemm_rcr_bias_add_77_0,

        reshape_89_0,
        global_workspace_,

     1,


        &reshape_87_0_dim_0,

        &reshape_87_0_dim_1,


        &encoder_layers_5_mha_proj_weight_dim_0,

        &encoder_layers_5_mha_proj_weight_dim_1,


        &reshape_87_0_dim_0,

        &encoder_layers_5_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_88" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_90" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_89_0_dim_0;

        M *= reshape_89_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_89_0_dim_2;

    
        layernorm_15(
           layernorm_90_0, reshape_89_0, encoder_layers_5_ln2_weight, encoder_layers_5_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_90" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_91" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_90_0,
        encoder_layers_5_mlp_fc1_weight,

        encoder_layers_5_mlp_fc1_bias,

        gemm_rcr_bias_gelu_91_0,
        global_workspace_,
        1,

        &reshape_89_0_dim_0,

        &reshape_89_0_dim_1,

        &reshape_89_0_dim_2,


        &encoder_layers_5_mlp_fc1_weight_dim_0,

        &encoder_layers_5_mlp_fc1_weight_dim_1,


        &reshape_89_0_dim_0,

        &reshape_89_0_dim_1,

        &encoder_layers_5_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_91" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_92" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_91_0,
        encoder_layers_5_mlp_fc2_weight,
        encoder_layers_5_mlp_fc2_bias,
        reshape_89_0,

        gemm_rcr_bias_add_92_0,
        global_workspace_,

     1,


        &reshape_89_0_dim_0,

        &reshape_89_0_dim_1,

        &encoder_layers_5_mlp_fc1_weight_dim_0,


        &encoder_layers_5_mlp_fc2_weight_dim_0,

        &encoder_layers_5_mlp_fc2_weight_dim_1,


        &reshape_89_0_dim_0,

        &reshape_89_0_dim_1,

        &encoder_layers_5_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_92" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_93" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_89_0_dim_0;

        M *= reshape_89_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_5_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_94_0, gemm_rcr_bias_add_92_0, encoder_layers_6_ln1_weight, encoder_layers_6_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_93" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_95" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_94_0,
        encoder_layers_6_mha_qkv_weight,

        encoder_layers_6_mha_qkv_bias,

        reshape_96_0,
        global_workspace_,
        1,

        &reshape_94_0_dim_0,

        &reshape_94_0_dim_1,


        &encoder_layers_6_mha_qkv_weight_dim_0,

        &encoder_layers_6_mha_qkv_weight_dim_1,


        &reshape_94_0_dim_0,

        &encoder_layers_6_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_95" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_97" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_97_2,
      split_97_0,
      split_97_1
    };


      int64_t *split_97_2_shape[] = {
        &split_97_2_dim_0, &reshape_96_0_dim_1, &reshape_96_0_dim_2, &reshape_96_0_dim_3, &reshape_96_0_dim_4
      };
      int64_t *split_97_0_shape[] = {
        &split_97_0_dim_0, &reshape_96_0_dim_1, &reshape_96_0_dim_2, &reshape_96_0_dim_3, &reshape_96_0_dim_4
      };
      int64_t *split_97_1_shape[] = {
        &split_97_1_dim_0, &reshape_96_0_dim_1, &reshape_96_0_dim_2, &reshape_96_0_dim_3, &reshape_96_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_97_2_shape, split_97_0_shape, split_97_1_shape
    };

    const int64_t reshape_96_0_shape[] = {
      reshape_96_0_dim_0, reshape_96_0_dim_1, reshape_96_0_dim_2, reshape_96_0_dim_3, reshape_96_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_96_0,
        reshape_96_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_97" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_98" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_97_0_dim_0,
        &reshape_96_0_dim_1,
        &reshape_96_0_dim_2,
        &reshape_96_0_dim_3,
        &reshape_96_0_dim_4,
        &reshape_98_0_dim_0,
        &reshape_98_0_dim_1,
        &reshape_98_0_dim_2,
        &reshape_98_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_98" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_99" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_97_1_dim_0,
        &reshape_96_0_dim_1,
        &reshape_96_0_dim_2,
        &reshape_96_0_dim_3,
        &reshape_96_0_dim_4,
        &reshape_99_0_dim_0,
        &reshape_99_0_dim_1,
        &reshape_99_0_dim_2,
        &reshape_99_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_99" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_100" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_97_2_dim_0,
        &reshape_96_0_dim_1,
        &reshape_96_0_dim_2,
        &reshape_96_0_dim_3,
        &reshape_96_0_dim_4,
        &reshape_100_0_dim_0,
        &reshape_100_0_dim_1,
        &reshape_100_0_dim_2,
        &reshape_100_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_100" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_101" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_101_0,
        reshape_100_0, reshape_98_0, reshape_99_0,
        &reshape_100_0_dim_0,
        &reshape_99_0_dim_2,
        &reshape_100_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_101" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_103" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_101_0,
        encoder_layers_6_mha_proj_weight,
        encoder_layers_6_mha_proj_bias,
        gemm_rcr_bias_add_92_0,

        reshape_104_0,
        global_workspace_,

     1,


        &reshape_102_0_dim_0,

        &reshape_102_0_dim_1,


        &encoder_layers_6_mha_proj_weight_dim_0,

        &encoder_layers_6_mha_proj_weight_dim_1,


        &reshape_102_0_dim_0,

        &encoder_layers_6_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_103" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_105" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_104_0_dim_0;

        M *= reshape_104_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_104_0_dim_2;

    
        layernorm_15(
           layernorm_105_0, reshape_104_0, encoder_layers_6_ln2_weight, encoder_layers_6_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_105" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_106" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_105_0,
        encoder_layers_6_mlp_fc1_weight,

        encoder_layers_6_mlp_fc1_bias,

        gemm_rcr_bias_gelu_106_0,
        global_workspace_,
        1,

        &reshape_104_0_dim_0,

        &reshape_104_0_dim_1,

        &reshape_104_0_dim_2,


        &encoder_layers_6_mlp_fc1_weight_dim_0,

        &encoder_layers_6_mlp_fc1_weight_dim_1,


        &reshape_104_0_dim_0,

        &reshape_104_0_dim_1,

        &encoder_layers_6_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_106" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_107" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_106_0,
        encoder_layers_6_mlp_fc2_weight,
        encoder_layers_6_mlp_fc2_bias,
        reshape_104_0,

        gemm_rcr_bias_add_107_0,
        global_workspace_,

     1,


        &reshape_104_0_dim_0,

        &reshape_104_0_dim_1,

        &encoder_layers_6_mlp_fc1_weight_dim_0,


        &encoder_layers_6_mlp_fc2_weight_dim_0,

        &encoder_layers_6_mlp_fc2_weight_dim_1,


        &reshape_104_0_dim_0,

        &reshape_104_0_dim_1,

        &encoder_layers_6_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_107" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_108" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_104_0_dim_0;

        M *= reshape_104_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_6_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_109_0, gemm_rcr_bias_add_107_0, encoder_layers_7_ln1_weight, encoder_layers_7_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_108" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_110" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_109_0,
        encoder_layers_7_mha_qkv_weight,

        encoder_layers_7_mha_qkv_bias,

        reshape_111_0,
        global_workspace_,
        1,

        &reshape_109_0_dim_0,

        &reshape_109_0_dim_1,


        &encoder_layers_7_mha_qkv_weight_dim_0,

        &encoder_layers_7_mha_qkv_weight_dim_1,


        &reshape_109_0_dim_0,

        &encoder_layers_7_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_110" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_112" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_112_0,
      split_112_1,
      split_112_2
    };


      int64_t *split_112_0_shape[] = {
        &split_112_0_dim_0, &reshape_111_0_dim_1, &reshape_111_0_dim_2, &reshape_111_0_dim_3, &reshape_111_0_dim_4
      };
      int64_t *split_112_1_shape[] = {
        &split_112_1_dim_0, &reshape_111_0_dim_1, &reshape_111_0_dim_2, &reshape_111_0_dim_3, &reshape_111_0_dim_4
      };
      int64_t *split_112_2_shape[] = {
        &split_112_2_dim_0, &reshape_111_0_dim_1, &reshape_111_0_dim_2, &reshape_111_0_dim_3, &reshape_111_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_112_0_shape, split_112_1_shape, split_112_2_shape
    };

    const int64_t reshape_111_0_shape[] = {
      reshape_111_0_dim_0, reshape_111_0_dim_1, reshape_111_0_dim_2, reshape_111_0_dim_3, reshape_111_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_111_0,
        reshape_111_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_112" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_113" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_112_0_dim_0,
        &reshape_111_0_dim_1,
        &reshape_111_0_dim_2,
        &reshape_111_0_dim_3,
        &reshape_111_0_dim_4,
        &reshape_113_0_dim_0,
        &reshape_113_0_dim_1,
        &reshape_113_0_dim_2,
        &reshape_113_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_113" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_114" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_112_1_dim_0,
        &reshape_111_0_dim_1,
        &reshape_111_0_dim_2,
        &reshape_111_0_dim_3,
        &reshape_111_0_dim_4,
        &reshape_114_0_dim_0,
        &reshape_114_0_dim_1,
        &reshape_114_0_dim_2,
        &reshape_114_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_114" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_115" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_112_2_dim_0,
        &reshape_111_0_dim_1,
        &reshape_111_0_dim_2,
        &reshape_111_0_dim_3,
        &reshape_111_0_dim_4,
        &reshape_115_0_dim_0,
        &reshape_115_0_dim_1,
        &reshape_115_0_dim_2,
        &reshape_115_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_115" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_116" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_116_0,
        reshape_113_0, reshape_114_0, reshape_115_0,
        &reshape_113_0_dim_0,
        &reshape_115_0_dim_2,
        &reshape_113_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_116" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_118" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_116_0,
        encoder_layers_7_mha_proj_weight,
        encoder_layers_7_mha_proj_bias,
        gemm_rcr_bias_add_107_0,

        reshape_119_0,
        global_workspace_,

     1,


        &reshape_117_0_dim_0,

        &reshape_117_0_dim_1,


        &encoder_layers_7_mha_proj_weight_dim_0,

        &encoder_layers_7_mha_proj_weight_dim_1,


        &reshape_117_0_dim_0,

        &encoder_layers_7_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_118" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_120" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_119_0_dim_0;

        M *= reshape_119_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_119_0_dim_2;

    
        layernorm_15(
           layernorm_120_0, reshape_119_0, encoder_layers_7_ln2_weight, encoder_layers_7_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_120" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_121" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_120_0,
        encoder_layers_7_mlp_fc1_weight,

        encoder_layers_7_mlp_fc1_bias,

        gemm_rcr_bias_gelu_121_0,
        global_workspace_,
        1,

        &reshape_119_0_dim_0,

        &reshape_119_0_dim_1,

        &reshape_119_0_dim_2,


        &encoder_layers_7_mlp_fc1_weight_dim_0,

        &encoder_layers_7_mlp_fc1_weight_dim_1,


        &reshape_119_0_dim_0,

        &reshape_119_0_dim_1,

        &encoder_layers_7_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_121" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_122" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_121_0,
        encoder_layers_7_mlp_fc2_weight,
        encoder_layers_7_mlp_fc2_bias,
        reshape_119_0,

        gemm_rcr_bias_add_122_0,
        global_workspace_,

     1,


        &reshape_119_0_dim_0,

        &reshape_119_0_dim_1,

        &encoder_layers_7_mlp_fc1_weight_dim_0,


        &encoder_layers_7_mlp_fc2_weight_dim_0,

        &encoder_layers_7_mlp_fc2_weight_dim_1,


        &reshape_119_0_dim_0,

        &reshape_119_0_dim_1,

        &encoder_layers_7_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_122" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_123" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_119_0_dim_0;

        M *= reshape_119_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_7_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_124_0, gemm_rcr_bias_add_122_0, encoder_layers_8_ln1_weight, encoder_layers_8_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_123" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_125" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_124_0,
        encoder_layers_8_mha_qkv_weight,

        encoder_layers_8_mha_qkv_bias,

        reshape_126_0,
        global_workspace_,
        1,

        &reshape_124_0_dim_0,

        &reshape_124_0_dim_1,


        &encoder_layers_8_mha_qkv_weight_dim_0,

        &encoder_layers_8_mha_qkv_weight_dim_1,


        &reshape_124_0_dim_0,

        &encoder_layers_8_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_125" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_127" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_127_1,
      split_127_0,
      split_127_2
    };


      int64_t *split_127_1_shape[] = {
        &split_127_1_dim_0, &reshape_126_0_dim_1, &reshape_126_0_dim_2, &reshape_126_0_dim_3, &reshape_126_0_dim_4
      };
      int64_t *split_127_0_shape[] = {
        &split_127_0_dim_0, &reshape_126_0_dim_1, &reshape_126_0_dim_2, &reshape_126_0_dim_3, &reshape_126_0_dim_4
      };
      int64_t *split_127_2_shape[] = {
        &split_127_2_dim_0, &reshape_126_0_dim_1, &reshape_126_0_dim_2, &reshape_126_0_dim_3, &reshape_126_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_127_1_shape, split_127_0_shape, split_127_2_shape
    };

    const int64_t reshape_126_0_shape[] = {
      reshape_126_0_dim_0, reshape_126_0_dim_1, reshape_126_0_dim_2, reshape_126_0_dim_3, reshape_126_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_126_0,
        reshape_126_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_127" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_128" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_127_0_dim_0,
        &reshape_126_0_dim_1,
        &reshape_126_0_dim_2,
        &reshape_126_0_dim_3,
        &reshape_126_0_dim_4,
        &reshape_128_0_dim_0,
        &reshape_128_0_dim_1,
        &reshape_128_0_dim_2,
        &reshape_128_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_128" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_129" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_127_1_dim_0,
        &reshape_126_0_dim_1,
        &reshape_126_0_dim_2,
        &reshape_126_0_dim_3,
        &reshape_126_0_dim_4,
        &reshape_129_0_dim_0,
        &reshape_129_0_dim_1,
        &reshape_129_0_dim_2,
        &reshape_129_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_129" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_130" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_127_2_dim_0,
        &reshape_126_0_dim_1,
        &reshape_126_0_dim_2,
        &reshape_126_0_dim_3,
        &reshape_126_0_dim_4,
        &reshape_130_0_dim_0,
        &reshape_130_0_dim_1,
        &reshape_130_0_dim_2,
        &reshape_130_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_130" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_131" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_131_0,
        reshape_129_0, reshape_128_0, reshape_130_0,
        &reshape_129_0_dim_0,
        &reshape_130_0_dim_2,
        &reshape_129_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_131" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_133" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_131_0,
        encoder_layers_8_mha_proj_weight,
        encoder_layers_8_mha_proj_bias,
        gemm_rcr_bias_add_122_0,

        reshape_134_0,
        global_workspace_,

     1,


        &reshape_132_0_dim_0,

        &reshape_132_0_dim_1,


        &encoder_layers_8_mha_proj_weight_dim_0,

        &encoder_layers_8_mha_proj_weight_dim_1,


        &reshape_132_0_dim_0,

        &encoder_layers_8_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_133" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_135" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_134_0_dim_0;

        M *= reshape_134_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_134_0_dim_2;

    
        layernorm_15(
           layernorm_135_0, reshape_134_0, encoder_layers_8_ln2_weight, encoder_layers_8_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_135" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_136" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_135_0,
        encoder_layers_8_mlp_fc1_weight,

        encoder_layers_8_mlp_fc1_bias,

        gemm_rcr_bias_gelu_136_0,
        global_workspace_,
        1,

        &reshape_134_0_dim_0,

        &reshape_134_0_dim_1,

        &reshape_134_0_dim_2,


        &encoder_layers_8_mlp_fc1_weight_dim_0,

        &encoder_layers_8_mlp_fc1_weight_dim_1,


        &reshape_134_0_dim_0,

        &reshape_134_0_dim_1,

        &encoder_layers_8_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_136" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_137" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_136_0,
        encoder_layers_8_mlp_fc2_weight,
        encoder_layers_8_mlp_fc2_bias,
        reshape_134_0,

        gemm_rcr_bias_add_137_0,
        global_workspace_,

     1,


        &reshape_134_0_dim_0,

        &reshape_134_0_dim_1,

        &encoder_layers_8_mlp_fc1_weight_dim_0,


        &encoder_layers_8_mlp_fc2_weight_dim_0,

        &encoder_layers_8_mlp_fc2_weight_dim_1,


        &reshape_134_0_dim_0,

        &reshape_134_0_dim_1,

        &encoder_layers_8_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_137" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_138" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_134_0_dim_0;

        M *= reshape_134_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_8_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_139_0, gemm_rcr_bias_add_137_0, encoder_layers_9_ln1_weight, encoder_layers_9_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_138" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_140" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_139_0,
        encoder_layers_9_mha_qkv_weight,

        encoder_layers_9_mha_qkv_bias,

        reshape_141_0,
        global_workspace_,
        1,

        &reshape_139_0_dim_0,

        &reshape_139_0_dim_1,


        &encoder_layers_9_mha_qkv_weight_dim_0,

        &encoder_layers_9_mha_qkv_weight_dim_1,


        &reshape_139_0_dim_0,

        &encoder_layers_9_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_140" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_142" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_142_2,
      split_142_1,
      split_142_0
    };


      int64_t *split_142_2_shape[] = {
        &split_142_2_dim_0, &reshape_141_0_dim_1, &reshape_141_0_dim_2, &reshape_141_0_dim_3, &reshape_141_0_dim_4
      };
      int64_t *split_142_1_shape[] = {
        &split_142_1_dim_0, &reshape_141_0_dim_1, &reshape_141_0_dim_2, &reshape_141_0_dim_3, &reshape_141_0_dim_4
      };
      int64_t *split_142_0_shape[] = {
        &split_142_0_dim_0, &reshape_141_0_dim_1, &reshape_141_0_dim_2, &reshape_141_0_dim_3, &reshape_141_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_142_2_shape, split_142_1_shape, split_142_0_shape
    };

    const int64_t reshape_141_0_shape[] = {
      reshape_141_0_dim_0, reshape_141_0_dim_1, reshape_141_0_dim_2, reshape_141_0_dim_3, reshape_141_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_141_0,
        reshape_141_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_142" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_143" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_142_0_dim_0,
        &reshape_141_0_dim_1,
        &reshape_141_0_dim_2,
        &reshape_141_0_dim_3,
        &reshape_141_0_dim_4,
        &reshape_143_0_dim_0,
        &reshape_143_0_dim_1,
        &reshape_143_0_dim_2,
        &reshape_143_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_143" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_144" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_142_1_dim_0,
        &reshape_141_0_dim_1,
        &reshape_141_0_dim_2,
        &reshape_141_0_dim_3,
        &reshape_141_0_dim_4,
        &reshape_144_0_dim_0,
        &reshape_144_0_dim_1,
        &reshape_144_0_dim_2,
        &reshape_144_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_144" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_145" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_142_2_dim_0,
        &reshape_141_0_dim_1,
        &reshape_141_0_dim_2,
        &reshape_141_0_dim_3,
        &reshape_141_0_dim_4,
        &reshape_145_0_dim_0,
        &reshape_145_0_dim_1,
        &reshape_145_0_dim_2,
        &reshape_145_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_145" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_146" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_146_0,
        reshape_145_0, reshape_144_0, reshape_143_0,
        &reshape_145_0_dim_0,
        &reshape_143_0_dim_2,
        &reshape_145_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_146" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_148" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_146_0,
        encoder_layers_9_mha_proj_weight,
        encoder_layers_9_mha_proj_bias,
        gemm_rcr_bias_add_137_0,

        reshape_149_0,
        global_workspace_,

     1,


        &reshape_147_0_dim_0,

        &reshape_147_0_dim_1,


        &encoder_layers_9_mha_proj_weight_dim_0,

        &encoder_layers_9_mha_proj_weight_dim_1,


        &reshape_147_0_dim_0,

        &encoder_layers_9_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_148" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_150" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_149_0_dim_0;

        M *= reshape_149_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_149_0_dim_2;

    
        layernorm_15(
           layernorm_150_0, reshape_149_0, encoder_layers_9_ln2_weight, encoder_layers_9_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_150" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_151" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_150_0,
        encoder_layers_9_mlp_fc1_weight,

        encoder_layers_9_mlp_fc1_bias,

        gemm_rcr_bias_gelu_151_0,
        global_workspace_,
        1,

        &reshape_149_0_dim_0,

        &reshape_149_0_dim_1,

        &reshape_149_0_dim_2,


        &encoder_layers_9_mlp_fc1_weight_dim_0,

        &encoder_layers_9_mlp_fc1_weight_dim_1,


        &reshape_149_0_dim_0,

        &reshape_149_0_dim_1,

        &encoder_layers_9_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_151" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_152" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_151_0,
        encoder_layers_9_mlp_fc2_weight,
        encoder_layers_9_mlp_fc2_bias,
        reshape_149_0,

        gemm_rcr_bias_add_152_0,
        global_workspace_,

     1,


        &reshape_149_0_dim_0,

        &reshape_149_0_dim_1,

        &encoder_layers_9_mlp_fc1_weight_dim_0,


        &encoder_layers_9_mlp_fc2_weight_dim_0,

        &encoder_layers_9_mlp_fc2_weight_dim_1,


        &reshape_149_0_dim_0,

        &reshape_149_0_dim_1,

        &encoder_layers_9_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_152" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_153" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_149_0_dim_0;

        M *= reshape_149_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_9_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_154_0, gemm_rcr_bias_add_152_0, encoder_layers_10_ln1_weight, encoder_layers_10_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_153" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_155" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_154_0,
        encoder_layers_10_mha_qkv_weight,

        encoder_layers_10_mha_qkv_bias,

        reshape_156_0,
        global_workspace_,
        1,

        &reshape_154_0_dim_0,

        &reshape_154_0_dim_1,


        &encoder_layers_10_mha_qkv_weight_dim_0,

        &encoder_layers_10_mha_qkv_weight_dim_1,


        &reshape_154_0_dim_0,

        &encoder_layers_10_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_155" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_157" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_157_1,
      split_157_2,
      split_157_0
    };


      int64_t *split_157_1_shape[] = {
        &split_157_1_dim_0, &reshape_156_0_dim_1, &reshape_156_0_dim_2, &reshape_156_0_dim_3, &reshape_156_0_dim_4
      };
      int64_t *split_157_2_shape[] = {
        &split_157_2_dim_0, &reshape_156_0_dim_1, &reshape_156_0_dim_2, &reshape_156_0_dim_3, &reshape_156_0_dim_4
      };
      int64_t *split_157_0_shape[] = {
        &split_157_0_dim_0, &reshape_156_0_dim_1, &reshape_156_0_dim_2, &reshape_156_0_dim_3, &reshape_156_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_157_1_shape, split_157_2_shape, split_157_0_shape
    };

    const int64_t reshape_156_0_shape[] = {
      reshape_156_0_dim_0, reshape_156_0_dim_1, reshape_156_0_dim_2, reshape_156_0_dim_3, reshape_156_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_156_0,
        reshape_156_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_157" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_158" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_157_0_dim_0,
        &reshape_156_0_dim_1,
        &reshape_156_0_dim_2,
        &reshape_156_0_dim_3,
        &reshape_156_0_dim_4,
        &reshape_158_0_dim_0,
        &reshape_158_0_dim_1,
        &reshape_158_0_dim_2,
        &reshape_158_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_158" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_159" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_157_1_dim_0,
        &reshape_156_0_dim_1,
        &reshape_156_0_dim_2,
        &reshape_156_0_dim_3,
        &reshape_156_0_dim_4,
        &reshape_159_0_dim_0,
        &reshape_159_0_dim_1,
        &reshape_159_0_dim_2,
        &reshape_159_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_159" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_160" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_157_2_dim_0,
        &reshape_156_0_dim_1,
        &reshape_156_0_dim_2,
        &reshape_156_0_dim_3,
        &reshape_156_0_dim_4,
        &reshape_160_0_dim_0,
        &reshape_160_0_dim_1,
        &reshape_160_0_dim_2,
        &reshape_160_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_160" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_161" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_161_0,
        reshape_159_0, reshape_160_0, reshape_158_0,
        &reshape_159_0_dim_0,
        &reshape_158_0_dim_2,
        &reshape_159_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_161" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_163" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_161_0,
        encoder_layers_10_mha_proj_weight,
        encoder_layers_10_mha_proj_bias,
        gemm_rcr_bias_add_152_0,

        reshape_164_0,
        global_workspace_,

     1,


        &reshape_162_0_dim_0,

        &reshape_162_0_dim_1,


        &encoder_layers_10_mha_proj_weight_dim_0,

        &encoder_layers_10_mha_proj_weight_dim_1,


        &reshape_162_0_dim_0,

        &encoder_layers_10_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_163" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_165" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_164_0_dim_0;

        M *= reshape_164_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_164_0_dim_2;

    
        layernorm_15(
           layernorm_165_0, reshape_164_0, encoder_layers_10_ln2_weight, encoder_layers_10_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_165" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_166" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_165_0,
        encoder_layers_10_mlp_fc1_weight,

        encoder_layers_10_mlp_fc1_bias,

        gemm_rcr_bias_gelu_166_0,
        global_workspace_,
        1,

        &reshape_164_0_dim_0,

        &reshape_164_0_dim_1,

        &reshape_164_0_dim_2,


        &encoder_layers_10_mlp_fc1_weight_dim_0,

        &encoder_layers_10_mlp_fc1_weight_dim_1,


        &reshape_164_0_dim_0,

        &reshape_164_0_dim_1,

        &encoder_layers_10_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_166" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_167" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_166_0,
        encoder_layers_10_mlp_fc2_weight,
        encoder_layers_10_mlp_fc2_bias,
        reshape_164_0,

        gemm_rcr_bias_add_167_0,
        global_workspace_,

     1,


        &reshape_164_0_dim_0,

        &reshape_164_0_dim_1,

        &encoder_layers_10_mlp_fc1_weight_dim_0,


        &encoder_layers_10_mlp_fc2_weight_dim_0,

        &encoder_layers_10_mlp_fc2_weight_dim_1,


        &reshape_164_0_dim_0,

        &reshape_164_0_dim_1,

        &encoder_layers_10_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_167" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_168" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_164_0_dim_0;

        M *= reshape_164_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_10_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_169_0, gemm_rcr_bias_add_167_0, encoder_layers_11_ln1_weight, encoder_layers_11_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_168" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_170" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_169_0,
        encoder_layers_11_mha_qkv_weight,

        encoder_layers_11_mha_qkv_bias,

        reshape_171_0,
        global_workspace_,
        1,

        &reshape_169_0_dim_0,

        &reshape_169_0_dim_1,


        &encoder_layers_11_mha_qkv_weight_dim_0,

        &encoder_layers_11_mha_qkv_weight_dim_1,


        &reshape_169_0_dim_0,

        &encoder_layers_11_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_170" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_172" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_172_0,
      split_172_1,
      split_172_2
    };


      int64_t *split_172_0_shape[] = {
        &split_172_0_dim_0, &reshape_171_0_dim_1, &reshape_171_0_dim_2, &reshape_171_0_dim_3, &reshape_171_0_dim_4
      };
      int64_t *split_172_1_shape[] = {
        &split_172_1_dim_0, &reshape_171_0_dim_1, &reshape_171_0_dim_2, &reshape_171_0_dim_3, &reshape_171_0_dim_4
      };
      int64_t *split_172_2_shape[] = {
        &split_172_2_dim_0, &reshape_171_0_dim_1, &reshape_171_0_dim_2, &reshape_171_0_dim_3, &reshape_171_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_172_0_shape, split_172_1_shape, split_172_2_shape
    };

    const int64_t reshape_171_0_shape[] = {
      reshape_171_0_dim_0, reshape_171_0_dim_1, reshape_171_0_dim_2, reshape_171_0_dim_3, reshape_171_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_171_0,
        reshape_171_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_172" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_173" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_172_0_dim_0,
        &reshape_171_0_dim_1,
        &reshape_171_0_dim_2,
        &reshape_171_0_dim_3,
        &reshape_171_0_dim_4,
        &reshape_173_0_dim_0,
        &reshape_173_0_dim_1,
        &reshape_173_0_dim_2,
        &reshape_173_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_173" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_174" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_172_1_dim_0,
        &reshape_171_0_dim_1,
        &reshape_171_0_dim_2,
        &reshape_171_0_dim_3,
        &reshape_171_0_dim_4,
        &reshape_174_0_dim_0,
        &reshape_174_0_dim_1,
        &reshape_174_0_dim_2,
        &reshape_174_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_174" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_175" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_172_2_dim_0,
        &reshape_171_0_dim_1,
        &reshape_171_0_dim_2,
        &reshape_171_0_dim_3,
        &reshape_171_0_dim_4,
        &reshape_175_0_dim_0,
        &reshape_175_0_dim_1,
        &reshape_175_0_dim_2,
        &reshape_175_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_175" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_176" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_176_0,
        reshape_173_0, reshape_174_0, reshape_175_0,
        &reshape_173_0_dim_0,
        &reshape_175_0_dim_2,
        &reshape_173_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_176" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_178" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_176_0,
        encoder_layers_11_mha_proj_weight,
        encoder_layers_11_mha_proj_bias,
        gemm_rcr_bias_add_167_0,

        reshape_179_0,
        global_workspace_,

     1,


        &reshape_177_0_dim_0,

        &reshape_177_0_dim_1,


        &encoder_layers_11_mha_proj_weight_dim_0,

        &encoder_layers_11_mha_proj_weight_dim_1,


        &reshape_177_0_dim_0,

        &encoder_layers_11_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_178" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_180" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_179_0_dim_0;

        M *= reshape_179_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_179_0_dim_2;

    
        layernorm_15(
           layernorm_180_0, reshape_179_0, encoder_layers_11_ln2_weight, encoder_layers_11_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_180" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_181" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_180_0,
        encoder_layers_11_mlp_fc1_weight,

        encoder_layers_11_mlp_fc1_bias,

        gemm_rcr_bias_gelu_181_0,
        global_workspace_,
        1,

        &reshape_179_0_dim_0,

        &reshape_179_0_dim_1,

        &reshape_179_0_dim_2,


        &encoder_layers_11_mlp_fc1_weight_dim_0,

        &encoder_layers_11_mlp_fc1_weight_dim_1,


        &reshape_179_0_dim_0,

        &reshape_179_0_dim_1,

        &encoder_layers_11_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_181" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_182" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_181_0,
        encoder_layers_11_mlp_fc2_weight,
        encoder_layers_11_mlp_fc2_bias,
        reshape_179_0,

        gemm_rcr_bias_add_182_0,
        global_workspace_,

     1,


        &reshape_179_0_dim_0,

        &reshape_179_0_dim_1,

        &encoder_layers_11_mlp_fc1_weight_dim_0,


        &encoder_layers_11_mlp_fc2_weight_dim_0,

        &encoder_layers_11_mlp_fc2_weight_dim_1,


        &reshape_179_0_dim_0,

        &reshape_179_0_dim_1,

        &encoder_layers_11_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_182" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_183" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_179_0_dim_0;

        M *= reshape_179_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_11_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_184_0, gemm_rcr_bias_add_182_0, encoder_layers_12_ln1_weight, encoder_layers_12_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_183" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_185" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_184_0,
        encoder_layers_12_mha_qkv_weight,

        encoder_layers_12_mha_qkv_bias,

        reshape_186_0,
        global_workspace_,
        1,

        &reshape_184_0_dim_0,

        &reshape_184_0_dim_1,


        &encoder_layers_12_mha_qkv_weight_dim_0,

        &encoder_layers_12_mha_qkv_weight_dim_1,


        &reshape_184_0_dim_0,

        &encoder_layers_12_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_185" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_187" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_187_2,
      split_187_0,
      split_187_1
    };


      int64_t *split_187_2_shape[] = {
        &split_187_2_dim_0, &reshape_186_0_dim_1, &reshape_186_0_dim_2, &reshape_186_0_dim_3, &reshape_186_0_dim_4
      };
      int64_t *split_187_0_shape[] = {
        &split_187_0_dim_0, &reshape_186_0_dim_1, &reshape_186_0_dim_2, &reshape_186_0_dim_3, &reshape_186_0_dim_4
      };
      int64_t *split_187_1_shape[] = {
        &split_187_1_dim_0, &reshape_186_0_dim_1, &reshape_186_0_dim_2, &reshape_186_0_dim_3, &reshape_186_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_187_2_shape, split_187_0_shape, split_187_1_shape
    };

    const int64_t reshape_186_0_shape[] = {
      reshape_186_0_dim_0, reshape_186_0_dim_1, reshape_186_0_dim_2, reshape_186_0_dim_3, reshape_186_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_186_0,
        reshape_186_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_187" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_188" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_187_0_dim_0,
        &reshape_186_0_dim_1,
        &reshape_186_0_dim_2,
        &reshape_186_0_dim_3,
        &reshape_186_0_dim_4,
        &reshape_188_0_dim_0,
        &reshape_188_0_dim_1,
        &reshape_188_0_dim_2,
        &reshape_188_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_188" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_189" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_187_1_dim_0,
        &reshape_186_0_dim_1,
        &reshape_186_0_dim_2,
        &reshape_186_0_dim_3,
        &reshape_186_0_dim_4,
        &reshape_189_0_dim_0,
        &reshape_189_0_dim_1,
        &reshape_189_0_dim_2,
        &reshape_189_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_189" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_190" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_187_2_dim_0,
        &reshape_186_0_dim_1,
        &reshape_186_0_dim_2,
        &reshape_186_0_dim_3,
        &reshape_186_0_dim_4,
        &reshape_190_0_dim_0,
        &reshape_190_0_dim_1,
        &reshape_190_0_dim_2,
        &reshape_190_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_190" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_191" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_191_0,
        reshape_190_0, reshape_188_0, reshape_189_0,
        &reshape_190_0_dim_0,
        &reshape_189_0_dim_2,
        &reshape_190_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_191" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_193" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_191_0,
        encoder_layers_12_mha_proj_weight,
        encoder_layers_12_mha_proj_bias,
        gemm_rcr_bias_add_182_0,

        reshape_194_0,
        global_workspace_,

     1,


        &reshape_192_0_dim_0,

        &reshape_192_0_dim_1,


        &encoder_layers_12_mha_proj_weight_dim_0,

        &encoder_layers_12_mha_proj_weight_dim_1,


        &reshape_192_0_dim_0,

        &encoder_layers_12_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_193" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_195" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_194_0_dim_0;

        M *= reshape_194_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_194_0_dim_2;

    
        layernorm_15(
           layernorm_195_0, reshape_194_0, encoder_layers_12_ln2_weight, encoder_layers_12_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_195" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_196" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_195_0,
        encoder_layers_12_mlp_fc1_weight,

        encoder_layers_12_mlp_fc1_bias,

        gemm_rcr_bias_gelu_196_0,
        global_workspace_,
        1,

        &reshape_194_0_dim_0,

        &reshape_194_0_dim_1,

        &reshape_194_0_dim_2,


        &encoder_layers_12_mlp_fc1_weight_dim_0,

        &encoder_layers_12_mlp_fc1_weight_dim_1,


        &reshape_194_0_dim_0,

        &reshape_194_0_dim_1,

        &encoder_layers_12_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_196" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_197" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_196_0,
        encoder_layers_12_mlp_fc2_weight,
        encoder_layers_12_mlp_fc2_bias,
        reshape_194_0,

        gemm_rcr_bias_add_197_0,
        global_workspace_,

     1,


        &reshape_194_0_dim_0,

        &reshape_194_0_dim_1,

        &encoder_layers_12_mlp_fc1_weight_dim_0,


        &encoder_layers_12_mlp_fc2_weight_dim_0,

        &encoder_layers_12_mlp_fc2_weight_dim_1,


        &reshape_194_0_dim_0,

        &reshape_194_0_dim_1,

        &encoder_layers_12_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_197" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_198" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_194_0_dim_0;

        M *= reshape_194_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_12_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_199_0, gemm_rcr_bias_add_197_0, encoder_layers_13_ln1_weight, encoder_layers_13_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_198" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_200" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_199_0,
        encoder_layers_13_mha_qkv_weight,

        encoder_layers_13_mha_qkv_bias,

        reshape_201_0,
        global_workspace_,
        1,

        &reshape_199_0_dim_0,

        &reshape_199_0_dim_1,


        &encoder_layers_13_mha_qkv_weight_dim_0,

        &encoder_layers_13_mha_qkv_weight_dim_1,


        &reshape_199_0_dim_0,

        &encoder_layers_13_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_200" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_202" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_202_2,
      split_202_1,
      split_202_0
    };


      int64_t *split_202_2_shape[] = {
        &split_202_2_dim_0, &reshape_201_0_dim_1, &reshape_201_0_dim_2, &reshape_201_0_dim_3, &reshape_201_0_dim_4
      };
      int64_t *split_202_1_shape[] = {
        &split_202_1_dim_0, &reshape_201_0_dim_1, &reshape_201_0_dim_2, &reshape_201_0_dim_3, &reshape_201_0_dim_4
      };
      int64_t *split_202_0_shape[] = {
        &split_202_0_dim_0, &reshape_201_0_dim_1, &reshape_201_0_dim_2, &reshape_201_0_dim_3, &reshape_201_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_202_2_shape, split_202_1_shape, split_202_0_shape
    };

    const int64_t reshape_201_0_shape[] = {
      reshape_201_0_dim_0, reshape_201_0_dim_1, reshape_201_0_dim_2, reshape_201_0_dim_3, reshape_201_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_201_0,
        reshape_201_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_202" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_203" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_202_0_dim_0,
        &reshape_201_0_dim_1,
        &reshape_201_0_dim_2,
        &reshape_201_0_dim_3,
        &reshape_201_0_dim_4,
        &reshape_203_0_dim_0,
        &reshape_203_0_dim_1,
        &reshape_203_0_dim_2,
        &reshape_203_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_203" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_204" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_202_1_dim_0,
        &reshape_201_0_dim_1,
        &reshape_201_0_dim_2,
        &reshape_201_0_dim_3,
        &reshape_201_0_dim_4,
        &reshape_204_0_dim_0,
        &reshape_204_0_dim_1,
        &reshape_204_0_dim_2,
        &reshape_204_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_204" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_205" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_202_2_dim_0,
        &reshape_201_0_dim_1,
        &reshape_201_0_dim_2,
        &reshape_201_0_dim_3,
        &reshape_201_0_dim_4,
        &reshape_205_0_dim_0,
        &reshape_205_0_dim_1,
        &reshape_205_0_dim_2,
        &reshape_205_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_205" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_206" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_206_0,
        reshape_205_0, reshape_204_0, reshape_203_0,
        &reshape_205_0_dim_0,
        &reshape_203_0_dim_2,
        &reshape_205_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_206" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_208" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_206_0,
        encoder_layers_13_mha_proj_weight,
        encoder_layers_13_mha_proj_bias,
        gemm_rcr_bias_add_197_0,

        reshape_209_0,
        global_workspace_,

     1,


        &reshape_207_0_dim_0,

        &reshape_207_0_dim_1,


        &encoder_layers_13_mha_proj_weight_dim_0,

        &encoder_layers_13_mha_proj_weight_dim_1,


        &reshape_207_0_dim_0,

        &encoder_layers_13_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_208" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_210" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_209_0_dim_0;

        M *= reshape_209_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_209_0_dim_2;

    
        layernorm_15(
           layernorm_210_0, reshape_209_0, encoder_layers_13_ln2_weight, encoder_layers_13_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_210" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_211" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_210_0,
        encoder_layers_13_mlp_fc1_weight,

        encoder_layers_13_mlp_fc1_bias,

        gemm_rcr_bias_gelu_211_0,
        global_workspace_,
        1,

        &reshape_209_0_dim_0,

        &reshape_209_0_dim_1,

        &reshape_209_0_dim_2,


        &encoder_layers_13_mlp_fc1_weight_dim_0,

        &encoder_layers_13_mlp_fc1_weight_dim_1,


        &reshape_209_0_dim_0,

        &reshape_209_0_dim_1,

        &encoder_layers_13_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_211" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_212" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_211_0,
        encoder_layers_13_mlp_fc2_weight,
        encoder_layers_13_mlp_fc2_bias,
        reshape_209_0,

        gemm_rcr_bias_add_212_0,
        global_workspace_,

     1,


        &reshape_209_0_dim_0,

        &reshape_209_0_dim_1,

        &encoder_layers_13_mlp_fc1_weight_dim_0,


        &encoder_layers_13_mlp_fc2_weight_dim_0,

        &encoder_layers_13_mlp_fc2_weight_dim_1,


        &reshape_209_0_dim_0,

        &reshape_209_0_dim_1,

        &encoder_layers_13_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_212" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_213" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_209_0_dim_0;

        M *= reshape_209_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_13_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_214_0, gemm_rcr_bias_add_212_0, encoder_layers_14_ln1_weight, encoder_layers_14_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_213" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_215" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_214_0,
        encoder_layers_14_mha_qkv_weight,

        encoder_layers_14_mha_qkv_bias,

        reshape_216_0,
        global_workspace_,
        1,

        &reshape_214_0_dim_0,

        &reshape_214_0_dim_1,


        &encoder_layers_14_mha_qkv_weight_dim_0,

        &encoder_layers_14_mha_qkv_weight_dim_1,


        &reshape_214_0_dim_0,

        &encoder_layers_14_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_215" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_217" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_217_2,
      split_217_0,
      split_217_1
    };


      int64_t *split_217_2_shape[] = {
        &split_217_2_dim_0, &reshape_216_0_dim_1, &reshape_216_0_dim_2, &reshape_216_0_dim_3, &reshape_216_0_dim_4
      };
      int64_t *split_217_0_shape[] = {
        &split_217_0_dim_0, &reshape_216_0_dim_1, &reshape_216_0_dim_2, &reshape_216_0_dim_3, &reshape_216_0_dim_4
      };
      int64_t *split_217_1_shape[] = {
        &split_217_1_dim_0, &reshape_216_0_dim_1, &reshape_216_0_dim_2, &reshape_216_0_dim_3, &reshape_216_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_217_2_shape, split_217_0_shape, split_217_1_shape
    };

    const int64_t reshape_216_0_shape[] = {
      reshape_216_0_dim_0, reshape_216_0_dim_1, reshape_216_0_dim_2, reshape_216_0_dim_3, reshape_216_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_216_0,
        reshape_216_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_217" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_218" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_217_0_dim_0,
        &reshape_216_0_dim_1,
        &reshape_216_0_dim_2,
        &reshape_216_0_dim_3,
        &reshape_216_0_dim_4,
        &reshape_218_0_dim_0,
        &reshape_218_0_dim_1,
        &reshape_218_0_dim_2,
        &reshape_218_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_218" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_219" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_217_1_dim_0,
        &reshape_216_0_dim_1,
        &reshape_216_0_dim_2,
        &reshape_216_0_dim_3,
        &reshape_216_0_dim_4,
        &reshape_219_0_dim_0,
        &reshape_219_0_dim_1,
        &reshape_219_0_dim_2,
        &reshape_219_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_219" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_220" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_217_2_dim_0,
        &reshape_216_0_dim_1,
        &reshape_216_0_dim_2,
        &reshape_216_0_dim_3,
        &reshape_216_0_dim_4,
        &reshape_220_0_dim_0,
        &reshape_220_0_dim_1,
        &reshape_220_0_dim_2,
        &reshape_220_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_220" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_221" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_221_0,
        reshape_220_0, reshape_218_0, reshape_219_0,
        &reshape_220_0_dim_0,
        &reshape_219_0_dim_2,
        &reshape_220_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_221" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_223" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_221_0,
        encoder_layers_14_mha_proj_weight,
        encoder_layers_14_mha_proj_bias,
        gemm_rcr_bias_add_212_0,

        reshape_224_0,
        global_workspace_,

     1,


        &reshape_222_0_dim_0,

        &reshape_222_0_dim_1,


        &encoder_layers_14_mha_proj_weight_dim_0,

        &encoder_layers_14_mha_proj_weight_dim_1,


        &reshape_222_0_dim_0,

        &encoder_layers_14_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_223" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_225" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_224_0_dim_0;

        M *= reshape_224_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_224_0_dim_2;

    
        layernorm_15(
           layernorm_225_0, reshape_224_0, encoder_layers_14_ln2_weight, encoder_layers_14_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_225" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_226" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_225_0,
        encoder_layers_14_mlp_fc1_weight,

        encoder_layers_14_mlp_fc1_bias,

        gemm_rcr_bias_gelu_226_0,
        global_workspace_,
        1,

        &reshape_224_0_dim_0,

        &reshape_224_0_dim_1,

        &reshape_224_0_dim_2,


        &encoder_layers_14_mlp_fc1_weight_dim_0,

        &encoder_layers_14_mlp_fc1_weight_dim_1,


        &reshape_224_0_dim_0,

        &reshape_224_0_dim_1,

        &encoder_layers_14_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_226" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_227" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_226_0,
        encoder_layers_14_mlp_fc2_weight,
        encoder_layers_14_mlp_fc2_bias,
        reshape_224_0,

        gemm_rcr_bias_add_227_0,
        global_workspace_,

     1,


        &reshape_224_0_dim_0,

        &reshape_224_0_dim_1,

        &encoder_layers_14_mlp_fc1_weight_dim_0,


        &encoder_layers_14_mlp_fc2_weight_dim_0,

        &encoder_layers_14_mlp_fc2_weight_dim_1,


        &reshape_224_0_dim_0,

        &reshape_224_0_dim_1,

        &encoder_layers_14_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_227" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_228" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_224_0_dim_0;

        M *= reshape_224_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_14_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_229_0, gemm_rcr_bias_add_227_0, encoder_layers_15_ln1_weight, encoder_layers_15_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_228" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_230" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_229_0,
        encoder_layers_15_mha_qkv_weight,

        encoder_layers_15_mha_qkv_bias,

        reshape_231_0,
        global_workspace_,
        1,

        &reshape_229_0_dim_0,

        &reshape_229_0_dim_1,


        &encoder_layers_15_mha_qkv_weight_dim_0,

        &encoder_layers_15_mha_qkv_weight_dim_1,


        &reshape_229_0_dim_0,

        &encoder_layers_15_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_230" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_232" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_232_0,
      split_232_1,
      split_232_2
    };


      int64_t *split_232_0_shape[] = {
        &split_232_0_dim_0, &reshape_231_0_dim_1, &reshape_231_0_dim_2, &reshape_231_0_dim_3, &reshape_231_0_dim_4
      };
      int64_t *split_232_1_shape[] = {
        &split_232_1_dim_0, &reshape_231_0_dim_1, &reshape_231_0_dim_2, &reshape_231_0_dim_3, &reshape_231_0_dim_4
      };
      int64_t *split_232_2_shape[] = {
        &split_232_2_dim_0, &reshape_231_0_dim_1, &reshape_231_0_dim_2, &reshape_231_0_dim_3, &reshape_231_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_232_0_shape, split_232_1_shape, split_232_2_shape
    };

    const int64_t reshape_231_0_shape[] = {
      reshape_231_0_dim_0, reshape_231_0_dim_1, reshape_231_0_dim_2, reshape_231_0_dim_3, reshape_231_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_231_0,
        reshape_231_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_232" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_233" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_232_0_dim_0,
        &reshape_231_0_dim_1,
        &reshape_231_0_dim_2,
        &reshape_231_0_dim_3,
        &reshape_231_0_dim_4,
        &reshape_233_0_dim_0,
        &reshape_233_0_dim_1,
        &reshape_233_0_dim_2,
        &reshape_233_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_233" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_234" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_232_1_dim_0,
        &reshape_231_0_dim_1,
        &reshape_231_0_dim_2,
        &reshape_231_0_dim_3,
        &reshape_231_0_dim_4,
        &reshape_234_0_dim_0,
        &reshape_234_0_dim_1,
        &reshape_234_0_dim_2,
        &reshape_234_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_234" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_235" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_232_2_dim_0,
        &reshape_231_0_dim_1,
        &reshape_231_0_dim_2,
        &reshape_231_0_dim_3,
        &reshape_231_0_dim_4,
        &reshape_235_0_dim_0,
        &reshape_235_0_dim_1,
        &reshape_235_0_dim_2,
        &reshape_235_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_235" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_236" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_236_0,
        reshape_233_0, reshape_234_0, reshape_235_0,
        &reshape_233_0_dim_0,
        &reshape_235_0_dim_2,
        &reshape_233_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_236" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_238" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_236_0,
        encoder_layers_15_mha_proj_weight,
        encoder_layers_15_mha_proj_bias,
        gemm_rcr_bias_add_227_0,

        reshape_239_0,
        global_workspace_,

     1,


        &reshape_237_0_dim_0,

        &reshape_237_0_dim_1,


        &encoder_layers_15_mha_proj_weight_dim_0,

        &encoder_layers_15_mha_proj_weight_dim_1,


        &reshape_237_0_dim_0,

        &encoder_layers_15_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_238" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_240" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_239_0_dim_0;

        M *= reshape_239_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_239_0_dim_2;

    
        layernorm_15(
           layernorm_240_0, reshape_239_0, encoder_layers_15_ln2_weight, encoder_layers_15_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_240" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_241" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_240_0,
        encoder_layers_15_mlp_fc1_weight,

        encoder_layers_15_mlp_fc1_bias,

        gemm_rcr_bias_gelu_241_0,
        global_workspace_,
        1,

        &reshape_239_0_dim_0,

        &reshape_239_0_dim_1,

        &reshape_239_0_dim_2,


        &encoder_layers_15_mlp_fc1_weight_dim_0,

        &encoder_layers_15_mlp_fc1_weight_dim_1,


        &reshape_239_0_dim_0,

        &reshape_239_0_dim_1,

        &encoder_layers_15_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_241" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_242" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_241_0,
        encoder_layers_15_mlp_fc2_weight,
        encoder_layers_15_mlp_fc2_bias,
        reshape_239_0,

        gemm_rcr_bias_add_242_0,
        global_workspace_,

     1,


        &reshape_239_0_dim_0,

        &reshape_239_0_dim_1,

        &encoder_layers_15_mlp_fc1_weight_dim_0,


        &encoder_layers_15_mlp_fc2_weight_dim_0,

        &encoder_layers_15_mlp_fc2_weight_dim_1,


        &reshape_239_0_dim_0,

        &reshape_239_0_dim_1,

        &encoder_layers_15_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_242" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_243" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_239_0_dim_0;

        M *= reshape_239_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_15_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_244_0, gemm_rcr_bias_add_242_0, encoder_layers_16_ln1_weight, encoder_layers_16_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_243" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_245" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_244_0,
        encoder_layers_16_mha_qkv_weight,

        encoder_layers_16_mha_qkv_bias,

        reshape_246_0,
        global_workspace_,
        1,

        &reshape_244_0_dim_0,

        &reshape_244_0_dim_1,


        &encoder_layers_16_mha_qkv_weight_dim_0,

        &encoder_layers_16_mha_qkv_weight_dim_1,


        &reshape_244_0_dim_0,

        &encoder_layers_16_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_245" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_247" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_247_1,
      split_247_0,
      split_247_2
    };


      int64_t *split_247_1_shape[] = {
        &split_247_1_dim_0, &reshape_246_0_dim_1, &reshape_246_0_dim_2, &reshape_246_0_dim_3, &reshape_246_0_dim_4
      };
      int64_t *split_247_0_shape[] = {
        &split_247_0_dim_0, &reshape_246_0_dim_1, &reshape_246_0_dim_2, &reshape_246_0_dim_3, &reshape_246_0_dim_4
      };
      int64_t *split_247_2_shape[] = {
        &split_247_2_dim_0, &reshape_246_0_dim_1, &reshape_246_0_dim_2, &reshape_246_0_dim_3, &reshape_246_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_247_1_shape, split_247_0_shape, split_247_2_shape
    };

    const int64_t reshape_246_0_shape[] = {
      reshape_246_0_dim_0, reshape_246_0_dim_1, reshape_246_0_dim_2, reshape_246_0_dim_3, reshape_246_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_246_0,
        reshape_246_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_247" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_248" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_247_0_dim_0,
        &reshape_246_0_dim_1,
        &reshape_246_0_dim_2,
        &reshape_246_0_dim_3,
        &reshape_246_0_dim_4,
        &reshape_248_0_dim_0,
        &reshape_248_0_dim_1,
        &reshape_248_0_dim_2,
        &reshape_248_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_248" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_249" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_247_1_dim_0,
        &reshape_246_0_dim_1,
        &reshape_246_0_dim_2,
        &reshape_246_0_dim_3,
        &reshape_246_0_dim_4,
        &reshape_249_0_dim_0,
        &reshape_249_0_dim_1,
        &reshape_249_0_dim_2,
        &reshape_249_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_249" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_250" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_247_2_dim_0,
        &reshape_246_0_dim_1,
        &reshape_246_0_dim_2,
        &reshape_246_0_dim_3,
        &reshape_246_0_dim_4,
        &reshape_250_0_dim_0,
        &reshape_250_0_dim_1,
        &reshape_250_0_dim_2,
        &reshape_250_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_250" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_251" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_251_0,
        reshape_249_0, reshape_248_0, reshape_250_0,
        &reshape_249_0_dim_0,
        &reshape_250_0_dim_2,
        &reshape_249_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_251" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_253" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_251_0,
        encoder_layers_16_mha_proj_weight,
        encoder_layers_16_mha_proj_bias,
        gemm_rcr_bias_add_242_0,

        reshape_254_0,
        global_workspace_,

     1,


        &reshape_252_0_dim_0,

        &reshape_252_0_dim_1,


        &encoder_layers_16_mha_proj_weight_dim_0,

        &encoder_layers_16_mha_proj_weight_dim_1,


        &reshape_252_0_dim_0,

        &encoder_layers_16_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_253" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_255" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_254_0_dim_0;

        M *= reshape_254_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_254_0_dim_2;

    
        layernorm_15(
           layernorm_255_0, reshape_254_0, encoder_layers_16_ln2_weight, encoder_layers_16_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_255" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_256" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_255_0,
        encoder_layers_16_mlp_fc1_weight,

        encoder_layers_16_mlp_fc1_bias,

        gemm_rcr_bias_gelu_256_0,
        global_workspace_,
        1,

        &reshape_254_0_dim_0,

        &reshape_254_0_dim_1,

        &reshape_254_0_dim_2,


        &encoder_layers_16_mlp_fc1_weight_dim_0,

        &encoder_layers_16_mlp_fc1_weight_dim_1,


        &reshape_254_0_dim_0,

        &reshape_254_0_dim_1,

        &encoder_layers_16_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_256" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_257" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_256_0,
        encoder_layers_16_mlp_fc2_weight,
        encoder_layers_16_mlp_fc2_bias,
        reshape_254_0,

        gemm_rcr_bias_add_257_0,
        global_workspace_,

     1,


        &reshape_254_0_dim_0,

        &reshape_254_0_dim_1,

        &encoder_layers_16_mlp_fc1_weight_dim_0,


        &encoder_layers_16_mlp_fc2_weight_dim_0,

        &encoder_layers_16_mlp_fc2_weight_dim_1,


        &reshape_254_0_dim_0,

        &reshape_254_0_dim_1,

        &encoder_layers_16_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_257" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_258" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_254_0_dim_0;

        M *= reshape_254_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_16_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_259_0, gemm_rcr_bias_add_257_0, encoder_layers_17_ln1_weight, encoder_layers_17_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_258" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_260" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_259_0,
        encoder_layers_17_mha_qkv_weight,

        encoder_layers_17_mha_qkv_bias,

        reshape_261_0,
        global_workspace_,
        1,

        &reshape_259_0_dim_0,

        &reshape_259_0_dim_1,


        &encoder_layers_17_mha_qkv_weight_dim_0,

        &encoder_layers_17_mha_qkv_weight_dim_1,


        &reshape_259_0_dim_0,

        &encoder_layers_17_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_260" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_262" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_262_0,
      split_262_1,
      split_262_2
    };


      int64_t *split_262_0_shape[] = {
        &split_262_0_dim_0, &reshape_261_0_dim_1, &reshape_261_0_dim_2, &reshape_261_0_dim_3, &reshape_261_0_dim_4
      };
      int64_t *split_262_1_shape[] = {
        &split_262_1_dim_0, &reshape_261_0_dim_1, &reshape_261_0_dim_2, &reshape_261_0_dim_3, &reshape_261_0_dim_4
      };
      int64_t *split_262_2_shape[] = {
        &split_262_2_dim_0, &reshape_261_0_dim_1, &reshape_261_0_dim_2, &reshape_261_0_dim_3, &reshape_261_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_262_0_shape, split_262_1_shape, split_262_2_shape
    };

    const int64_t reshape_261_0_shape[] = {
      reshape_261_0_dim_0, reshape_261_0_dim_1, reshape_261_0_dim_2, reshape_261_0_dim_3, reshape_261_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_261_0,
        reshape_261_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_262" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_263" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_262_0_dim_0,
        &reshape_261_0_dim_1,
        &reshape_261_0_dim_2,
        &reshape_261_0_dim_3,
        &reshape_261_0_dim_4,
        &reshape_263_0_dim_0,
        &reshape_263_0_dim_1,
        &reshape_263_0_dim_2,
        &reshape_263_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_263" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_264" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_262_1_dim_0,
        &reshape_261_0_dim_1,
        &reshape_261_0_dim_2,
        &reshape_261_0_dim_3,
        &reshape_261_0_dim_4,
        &reshape_264_0_dim_0,
        &reshape_264_0_dim_1,
        &reshape_264_0_dim_2,
        &reshape_264_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_264" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_265" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_262_2_dim_0,
        &reshape_261_0_dim_1,
        &reshape_261_0_dim_2,
        &reshape_261_0_dim_3,
        &reshape_261_0_dim_4,
        &reshape_265_0_dim_0,
        &reshape_265_0_dim_1,
        &reshape_265_0_dim_2,
        &reshape_265_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_265" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_266" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_266_0,
        reshape_263_0, reshape_264_0, reshape_265_0,
        &reshape_263_0_dim_0,
        &reshape_265_0_dim_2,
        &reshape_263_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_266" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_268" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_266_0,
        encoder_layers_17_mha_proj_weight,
        encoder_layers_17_mha_proj_bias,
        gemm_rcr_bias_add_257_0,

        reshape_269_0,
        global_workspace_,

     1,


        &reshape_267_0_dim_0,

        &reshape_267_0_dim_1,


        &encoder_layers_17_mha_proj_weight_dim_0,

        &encoder_layers_17_mha_proj_weight_dim_1,


        &reshape_267_0_dim_0,

        &encoder_layers_17_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_268" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_270" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_269_0_dim_0;

        M *= reshape_269_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_269_0_dim_2;

    
        layernorm_15(
           layernorm_270_0, reshape_269_0, encoder_layers_17_ln2_weight, encoder_layers_17_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_270" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_271" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_270_0,
        encoder_layers_17_mlp_fc1_weight,

        encoder_layers_17_mlp_fc1_bias,

        gemm_rcr_bias_gelu_271_0,
        global_workspace_,
        1,

        &reshape_269_0_dim_0,

        &reshape_269_0_dim_1,

        &reshape_269_0_dim_2,


        &encoder_layers_17_mlp_fc1_weight_dim_0,

        &encoder_layers_17_mlp_fc1_weight_dim_1,


        &reshape_269_0_dim_0,

        &reshape_269_0_dim_1,

        &encoder_layers_17_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_271" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_272" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_271_0,
        encoder_layers_17_mlp_fc2_weight,
        encoder_layers_17_mlp_fc2_bias,
        reshape_269_0,

        gemm_rcr_bias_add_272_0,
        global_workspace_,

     1,


        &reshape_269_0_dim_0,

        &reshape_269_0_dim_1,

        &encoder_layers_17_mlp_fc1_weight_dim_0,


        &encoder_layers_17_mlp_fc2_weight_dim_0,

        &encoder_layers_17_mlp_fc2_weight_dim_1,


        &reshape_269_0_dim_0,

        &reshape_269_0_dim_1,

        &encoder_layers_17_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_272" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_273" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_269_0_dim_0;

        M *= reshape_269_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_17_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_274_0, gemm_rcr_bias_add_272_0, encoder_layers_18_ln1_weight, encoder_layers_18_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_273" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_275" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_274_0,
        encoder_layers_18_mha_qkv_weight,

        encoder_layers_18_mha_qkv_bias,

        reshape_276_0,
        global_workspace_,
        1,

        &reshape_274_0_dim_0,

        &reshape_274_0_dim_1,


        &encoder_layers_18_mha_qkv_weight_dim_0,

        &encoder_layers_18_mha_qkv_weight_dim_1,


        &reshape_274_0_dim_0,

        &encoder_layers_18_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_275" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_277" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_277_1,
      split_277_0,
      split_277_2
    };


      int64_t *split_277_1_shape[] = {
        &split_277_1_dim_0, &reshape_276_0_dim_1, &reshape_276_0_dim_2, &reshape_276_0_dim_3, &reshape_276_0_dim_4
      };
      int64_t *split_277_0_shape[] = {
        &split_277_0_dim_0, &reshape_276_0_dim_1, &reshape_276_0_dim_2, &reshape_276_0_dim_3, &reshape_276_0_dim_4
      };
      int64_t *split_277_2_shape[] = {
        &split_277_2_dim_0, &reshape_276_0_dim_1, &reshape_276_0_dim_2, &reshape_276_0_dim_3, &reshape_276_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_277_1_shape, split_277_0_shape, split_277_2_shape
    };

    const int64_t reshape_276_0_shape[] = {
      reshape_276_0_dim_0, reshape_276_0_dim_1, reshape_276_0_dim_2, reshape_276_0_dim_3, reshape_276_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_276_0,
        reshape_276_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_277" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_278" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_277_0_dim_0,
        &reshape_276_0_dim_1,
        &reshape_276_0_dim_2,
        &reshape_276_0_dim_3,
        &reshape_276_0_dim_4,
        &reshape_278_0_dim_0,
        &reshape_278_0_dim_1,
        &reshape_278_0_dim_2,
        &reshape_278_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_278" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_279" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_277_1_dim_0,
        &reshape_276_0_dim_1,
        &reshape_276_0_dim_2,
        &reshape_276_0_dim_3,
        &reshape_276_0_dim_4,
        &reshape_279_0_dim_0,
        &reshape_279_0_dim_1,
        &reshape_279_0_dim_2,
        &reshape_279_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_279" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_280" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_277_2_dim_0,
        &reshape_276_0_dim_1,
        &reshape_276_0_dim_2,
        &reshape_276_0_dim_3,
        &reshape_276_0_dim_4,
        &reshape_280_0_dim_0,
        &reshape_280_0_dim_1,
        &reshape_280_0_dim_2,
        &reshape_280_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_280" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_281" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_281_0,
        reshape_279_0, reshape_278_0, reshape_280_0,
        &reshape_279_0_dim_0,
        &reshape_280_0_dim_2,
        &reshape_279_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_281" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_283" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_281_0,
        encoder_layers_18_mha_proj_weight,
        encoder_layers_18_mha_proj_bias,
        gemm_rcr_bias_add_272_0,

        reshape_284_0,
        global_workspace_,

     1,


        &reshape_282_0_dim_0,

        &reshape_282_0_dim_1,


        &encoder_layers_18_mha_proj_weight_dim_0,

        &encoder_layers_18_mha_proj_weight_dim_1,


        &reshape_282_0_dim_0,

        &encoder_layers_18_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_283" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_285" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_284_0_dim_0;

        M *= reshape_284_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_284_0_dim_2;

    
        layernorm_15(
           layernorm_285_0, reshape_284_0, encoder_layers_18_ln2_weight, encoder_layers_18_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_285" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_286" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_285_0,
        encoder_layers_18_mlp_fc1_weight,

        encoder_layers_18_mlp_fc1_bias,

        gemm_rcr_bias_gelu_286_0,
        global_workspace_,
        1,

        &reshape_284_0_dim_0,

        &reshape_284_0_dim_1,

        &reshape_284_0_dim_2,


        &encoder_layers_18_mlp_fc1_weight_dim_0,

        &encoder_layers_18_mlp_fc1_weight_dim_1,


        &reshape_284_0_dim_0,

        &reshape_284_0_dim_1,

        &encoder_layers_18_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_286" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_287" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_286_0,
        encoder_layers_18_mlp_fc2_weight,
        encoder_layers_18_mlp_fc2_bias,
        reshape_284_0,

        gemm_rcr_bias_add_287_0,
        global_workspace_,

     1,


        &reshape_284_0_dim_0,

        &reshape_284_0_dim_1,

        &encoder_layers_18_mlp_fc1_weight_dim_0,


        &encoder_layers_18_mlp_fc2_weight_dim_0,

        &encoder_layers_18_mlp_fc2_weight_dim_1,


        &reshape_284_0_dim_0,

        &reshape_284_0_dim_1,

        &encoder_layers_18_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_287" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_288" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_284_0_dim_0;

        M *= reshape_284_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_18_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_289_0, gemm_rcr_bias_add_287_0, encoder_layers_19_ln1_weight, encoder_layers_19_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_288" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_290" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_289_0,
        encoder_layers_19_mha_qkv_weight,

        encoder_layers_19_mha_qkv_bias,

        reshape_291_0,
        global_workspace_,
        1,

        &reshape_289_0_dim_0,

        &reshape_289_0_dim_1,


        &encoder_layers_19_mha_qkv_weight_dim_0,

        &encoder_layers_19_mha_qkv_weight_dim_1,


        &reshape_289_0_dim_0,

        &encoder_layers_19_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_290" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_292" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_292_0,
      split_292_1,
      split_292_2
    };


      int64_t *split_292_0_shape[] = {
        &split_292_0_dim_0, &reshape_291_0_dim_1, &reshape_291_0_dim_2, &reshape_291_0_dim_3, &reshape_291_0_dim_4
      };
      int64_t *split_292_1_shape[] = {
        &split_292_1_dim_0, &reshape_291_0_dim_1, &reshape_291_0_dim_2, &reshape_291_0_dim_3, &reshape_291_0_dim_4
      };
      int64_t *split_292_2_shape[] = {
        &split_292_2_dim_0, &reshape_291_0_dim_1, &reshape_291_0_dim_2, &reshape_291_0_dim_3, &reshape_291_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_292_0_shape, split_292_1_shape, split_292_2_shape
    };

    const int64_t reshape_291_0_shape[] = {
      reshape_291_0_dim_0, reshape_291_0_dim_1, reshape_291_0_dim_2, reshape_291_0_dim_3, reshape_291_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_291_0,
        reshape_291_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_292" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_293" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_292_0_dim_0,
        &reshape_291_0_dim_1,
        &reshape_291_0_dim_2,
        &reshape_291_0_dim_3,
        &reshape_291_0_dim_4,
        &reshape_293_0_dim_0,
        &reshape_293_0_dim_1,
        &reshape_293_0_dim_2,
        &reshape_293_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_293" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_294" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_292_1_dim_0,
        &reshape_291_0_dim_1,
        &reshape_291_0_dim_2,
        &reshape_291_0_dim_3,
        &reshape_291_0_dim_4,
        &reshape_294_0_dim_0,
        &reshape_294_0_dim_1,
        &reshape_294_0_dim_2,
        &reshape_294_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_294" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_295" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_292_2_dim_0,
        &reshape_291_0_dim_1,
        &reshape_291_0_dim_2,
        &reshape_291_0_dim_3,
        &reshape_291_0_dim_4,
        &reshape_295_0_dim_0,
        &reshape_295_0_dim_1,
        &reshape_295_0_dim_2,
        &reshape_295_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_295" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_296" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_296_0,
        reshape_293_0, reshape_294_0, reshape_295_0,
        &reshape_293_0_dim_0,
        &reshape_295_0_dim_2,
        &reshape_293_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_296" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_298" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_296_0,
        encoder_layers_19_mha_proj_weight,
        encoder_layers_19_mha_proj_bias,
        gemm_rcr_bias_add_287_0,

        reshape_299_0,
        global_workspace_,

     1,


        &reshape_297_0_dim_0,

        &reshape_297_0_dim_1,


        &encoder_layers_19_mha_proj_weight_dim_0,

        &encoder_layers_19_mha_proj_weight_dim_1,


        &reshape_297_0_dim_0,

        &encoder_layers_19_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_298" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_300" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_299_0_dim_0;

        M *= reshape_299_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_299_0_dim_2;

    
        layernorm_15(
           layernorm_300_0, reshape_299_0, encoder_layers_19_ln2_weight, encoder_layers_19_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_300" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_301" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_300_0,
        encoder_layers_19_mlp_fc1_weight,

        encoder_layers_19_mlp_fc1_bias,

        gemm_rcr_bias_gelu_301_0,
        global_workspace_,
        1,

        &reshape_299_0_dim_0,

        &reshape_299_0_dim_1,

        &reshape_299_0_dim_2,


        &encoder_layers_19_mlp_fc1_weight_dim_0,

        &encoder_layers_19_mlp_fc1_weight_dim_1,


        &reshape_299_0_dim_0,

        &reshape_299_0_dim_1,

        &encoder_layers_19_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_301" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_302" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_301_0,
        encoder_layers_19_mlp_fc2_weight,
        encoder_layers_19_mlp_fc2_bias,
        reshape_299_0,

        gemm_rcr_bias_add_302_0,
        global_workspace_,

     1,


        &reshape_299_0_dim_0,

        &reshape_299_0_dim_1,

        &encoder_layers_19_mlp_fc1_weight_dim_0,


        &encoder_layers_19_mlp_fc2_weight_dim_0,

        &encoder_layers_19_mlp_fc2_weight_dim_1,


        &reshape_299_0_dim_0,

        &reshape_299_0_dim_1,

        &encoder_layers_19_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_302" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_303" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_299_0_dim_0;

        M *= reshape_299_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_19_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_304_0, gemm_rcr_bias_add_302_0, encoder_layers_20_ln1_weight, encoder_layers_20_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_303" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_305" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_304_0,
        encoder_layers_20_mha_qkv_weight,

        encoder_layers_20_mha_qkv_bias,

        reshape_306_0,
        global_workspace_,
        1,

        &reshape_304_0_dim_0,

        &reshape_304_0_dim_1,


        &encoder_layers_20_mha_qkv_weight_dim_0,

        &encoder_layers_20_mha_qkv_weight_dim_1,


        &reshape_304_0_dim_0,

        &encoder_layers_20_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_305" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_307" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_307_2,
      split_307_1,
      split_307_0
    };


      int64_t *split_307_2_shape[] = {
        &split_307_2_dim_0, &reshape_306_0_dim_1, &reshape_306_0_dim_2, &reshape_306_0_dim_3, &reshape_306_0_dim_4
      };
      int64_t *split_307_1_shape[] = {
        &split_307_1_dim_0, &reshape_306_0_dim_1, &reshape_306_0_dim_2, &reshape_306_0_dim_3, &reshape_306_0_dim_4
      };
      int64_t *split_307_0_shape[] = {
        &split_307_0_dim_0, &reshape_306_0_dim_1, &reshape_306_0_dim_2, &reshape_306_0_dim_3, &reshape_306_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_307_2_shape, split_307_1_shape, split_307_0_shape
    };

    const int64_t reshape_306_0_shape[] = {
      reshape_306_0_dim_0, reshape_306_0_dim_1, reshape_306_0_dim_2, reshape_306_0_dim_3, reshape_306_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_306_0,
        reshape_306_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_307" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_308" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_307_0_dim_0,
        &reshape_306_0_dim_1,
        &reshape_306_0_dim_2,
        &reshape_306_0_dim_3,
        &reshape_306_0_dim_4,
        &reshape_308_0_dim_0,
        &reshape_308_0_dim_1,
        &reshape_308_0_dim_2,
        &reshape_308_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_308" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_309" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_307_1_dim_0,
        &reshape_306_0_dim_1,
        &reshape_306_0_dim_2,
        &reshape_306_0_dim_3,
        &reshape_306_0_dim_4,
        &reshape_309_0_dim_0,
        &reshape_309_0_dim_1,
        &reshape_309_0_dim_2,
        &reshape_309_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_309" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_310" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_307_2_dim_0,
        &reshape_306_0_dim_1,
        &reshape_306_0_dim_2,
        &reshape_306_0_dim_3,
        &reshape_306_0_dim_4,
        &reshape_310_0_dim_0,
        &reshape_310_0_dim_1,
        &reshape_310_0_dim_2,
        &reshape_310_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_310" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_311" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_311_0,
        reshape_310_0, reshape_309_0, reshape_308_0,
        &reshape_310_0_dim_0,
        &reshape_308_0_dim_2,
        &reshape_310_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_311" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_313" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_311_0,
        encoder_layers_20_mha_proj_weight,
        encoder_layers_20_mha_proj_bias,
        gemm_rcr_bias_add_302_0,

        reshape_314_0,
        global_workspace_,

     1,


        &reshape_312_0_dim_0,

        &reshape_312_0_dim_1,


        &encoder_layers_20_mha_proj_weight_dim_0,

        &encoder_layers_20_mha_proj_weight_dim_1,


        &reshape_312_0_dim_0,

        &encoder_layers_20_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_313" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_315" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_314_0_dim_0;

        M *= reshape_314_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_314_0_dim_2;

    
        layernorm_15(
           layernorm_315_0, reshape_314_0, encoder_layers_20_ln2_weight, encoder_layers_20_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_315" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_316" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_315_0,
        encoder_layers_20_mlp_fc1_weight,

        encoder_layers_20_mlp_fc1_bias,

        gemm_rcr_bias_gelu_316_0,
        global_workspace_,
        1,

        &reshape_314_0_dim_0,

        &reshape_314_0_dim_1,

        &reshape_314_0_dim_2,


        &encoder_layers_20_mlp_fc1_weight_dim_0,

        &encoder_layers_20_mlp_fc1_weight_dim_1,


        &reshape_314_0_dim_0,

        &reshape_314_0_dim_1,

        &encoder_layers_20_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_316" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_317" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_316_0,
        encoder_layers_20_mlp_fc2_weight,
        encoder_layers_20_mlp_fc2_bias,
        reshape_314_0,

        gemm_rcr_bias_add_317_0,
        global_workspace_,

     1,


        &reshape_314_0_dim_0,

        &reshape_314_0_dim_1,

        &encoder_layers_20_mlp_fc1_weight_dim_0,


        &encoder_layers_20_mlp_fc2_weight_dim_0,

        &encoder_layers_20_mlp_fc2_weight_dim_1,


        &reshape_314_0_dim_0,

        &reshape_314_0_dim_1,

        &encoder_layers_20_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_317" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_318" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_314_0_dim_0;

        M *= reshape_314_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_20_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_319_0, gemm_rcr_bias_add_317_0, encoder_layers_21_ln1_weight, encoder_layers_21_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_318" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_320" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_319_0,
        encoder_layers_21_mha_qkv_weight,

        encoder_layers_21_mha_qkv_bias,

        reshape_321_0,
        global_workspace_,
        1,

        &reshape_319_0_dim_0,

        &reshape_319_0_dim_1,


        &encoder_layers_21_mha_qkv_weight_dim_0,

        &encoder_layers_21_mha_qkv_weight_dim_1,


        &reshape_319_0_dim_0,

        &encoder_layers_21_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_320" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_322" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_322_1,
      split_322_2,
      split_322_0
    };


      int64_t *split_322_1_shape[] = {
        &split_322_1_dim_0, &reshape_321_0_dim_1, &reshape_321_0_dim_2, &reshape_321_0_dim_3, &reshape_321_0_dim_4
      };
      int64_t *split_322_2_shape[] = {
        &split_322_2_dim_0, &reshape_321_0_dim_1, &reshape_321_0_dim_2, &reshape_321_0_dim_3, &reshape_321_0_dim_4
      };
      int64_t *split_322_0_shape[] = {
        &split_322_0_dim_0, &reshape_321_0_dim_1, &reshape_321_0_dim_2, &reshape_321_0_dim_3, &reshape_321_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_322_1_shape, split_322_2_shape, split_322_0_shape
    };

    const int64_t reshape_321_0_shape[] = {
      reshape_321_0_dim_0, reshape_321_0_dim_1, reshape_321_0_dim_2, reshape_321_0_dim_3, reshape_321_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_321_0,
        reshape_321_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_322" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_323" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_322_0_dim_0,
        &reshape_321_0_dim_1,
        &reshape_321_0_dim_2,
        &reshape_321_0_dim_3,
        &reshape_321_0_dim_4,
        &reshape_323_0_dim_0,
        &reshape_323_0_dim_1,
        &reshape_323_0_dim_2,
        &reshape_323_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_323" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_324" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_322_1_dim_0,
        &reshape_321_0_dim_1,
        &reshape_321_0_dim_2,
        &reshape_321_0_dim_3,
        &reshape_321_0_dim_4,
        &reshape_324_0_dim_0,
        &reshape_324_0_dim_1,
        &reshape_324_0_dim_2,
        &reshape_324_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_324" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_325" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_322_2_dim_0,
        &reshape_321_0_dim_1,
        &reshape_321_0_dim_2,
        &reshape_321_0_dim_3,
        &reshape_321_0_dim_4,
        &reshape_325_0_dim_0,
        &reshape_325_0_dim_1,
        &reshape_325_0_dim_2,
        &reshape_325_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_325" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_326" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_326_0,
        reshape_324_0, reshape_325_0, reshape_323_0,
        &reshape_324_0_dim_0,
        &reshape_323_0_dim_2,
        &reshape_324_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_326" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_328" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_326_0,
        encoder_layers_21_mha_proj_weight,
        encoder_layers_21_mha_proj_bias,
        gemm_rcr_bias_add_317_0,

        reshape_329_0,
        global_workspace_,

     1,


        &reshape_327_0_dim_0,

        &reshape_327_0_dim_1,


        &encoder_layers_21_mha_proj_weight_dim_0,

        &encoder_layers_21_mha_proj_weight_dim_1,


        &reshape_327_0_dim_0,

        &encoder_layers_21_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_328" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_330" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_329_0_dim_0;

        M *= reshape_329_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_329_0_dim_2;

    
        layernorm_15(
           layernorm_330_0, reshape_329_0, encoder_layers_21_ln2_weight, encoder_layers_21_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_330" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_331" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_330_0,
        encoder_layers_21_mlp_fc1_weight,

        encoder_layers_21_mlp_fc1_bias,

        gemm_rcr_bias_gelu_331_0,
        global_workspace_,
        1,

        &reshape_329_0_dim_0,

        &reshape_329_0_dim_1,

        &reshape_329_0_dim_2,


        &encoder_layers_21_mlp_fc1_weight_dim_0,

        &encoder_layers_21_mlp_fc1_weight_dim_1,


        &reshape_329_0_dim_0,

        &reshape_329_0_dim_1,

        &encoder_layers_21_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_331" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_332" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_331_0,
        encoder_layers_21_mlp_fc2_weight,
        encoder_layers_21_mlp_fc2_bias,
        reshape_329_0,

        gemm_rcr_bias_add_332_0,
        global_workspace_,

     1,


        &reshape_329_0_dim_0,

        &reshape_329_0_dim_1,

        &encoder_layers_21_mlp_fc1_weight_dim_0,


        &encoder_layers_21_mlp_fc2_weight_dim_0,

        &encoder_layers_21_mlp_fc2_weight_dim_1,


        &reshape_329_0_dim_0,

        &reshape_329_0_dim_1,

        &encoder_layers_21_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_332" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_333" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_329_0_dim_0;

        M *= reshape_329_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_21_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_334_0, gemm_rcr_bias_add_332_0, encoder_layers_22_ln1_weight, encoder_layers_22_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_333" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_335" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_334_0,
        encoder_layers_22_mha_qkv_weight,

        encoder_layers_22_mha_qkv_bias,

        reshape_336_0,
        global_workspace_,
        1,

        &reshape_334_0_dim_0,

        &reshape_334_0_dim_1,


        &encoder_layers_22_mha_qkv_weight_dim_0,

        &encoder_layers_22_mha_qkv_weight_dim_1,


        &reshape_334_0_dim_0,

        &encoder_layers_22_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_335" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_337" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_337_0,
      split_337_1,
      split_337_2
    };


      int64_t *split_337_0_shape[] = {
        &split_337_0_dim_0, &reshape_336_0_dim_1, &reshape_336_0_dim_2, &reshape_336_0_dim_3, &reshape_336_0_dim_4
      };
      int64_t *split_337_1_shape[] = {
        &split_337_1_dim_0, &reshape_336_0_dim_1, &reshape_336_0_dim_2, &reshape_336_0_dim_3, &reshape_336_0_dim_4
      };
      int64_t *split_337_2_shape[] = {
        &split_337_2_dim_0, &reshape_336_0_dim_1, &reshape_336_0_dim_2, &reshape_336_0_dim_3, &reshape_336_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_337_0_shape, split_337_1_shape, split_337_2_shape
    };

    const int64_t reshape_336_0_shape[] = {
      reshape_336_0_dim_0, reshape_336_0_dim_1, reshape_336_0_dim_2, reshape_336_0_dim_3, reshape_336_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_336_0,
        reshape_336_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_337" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_338" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_337_0_dim_0,
        &reshape_336_0_dim_1,
        &reshape_336_0_dim_2,
        &reshape_336_0_dim_3,
        &reshape_336_0_dim_4,
        &reshape_338_0_dim_0,
        &reshape_338_0_dim_1,
        &reshape_338_0_dim_2,
        &reshape_338_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_338" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_339" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_337_1_dim_0,
        &reshape_336_0_dim_1,
        &reshape_336_0_dim_2,
        &reshape_336_0_dim_3,
        &reshape_336_0_dim_4,
        &reshape_339_0_dim_0,
        &reshape_339_0_dim_1,
        &reshape_339_0_dim_2,
        &reshape_339_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_339" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_340" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_337_2_dim_0,
        &reshape_336_0_dim_1,
        &reshape_336_0_dim_2,
        &reshape_336_0_dim_3,
        &reshape_336_0_dim_4,
        &reshape_340_0_dim_0,
        &reshape_340_0_dim_1,
        &reshape_340_0_dim_2,
        &reshape_340_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_340" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_341" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_341_0,
        reshape_338_0, reshape_339_0, reshape_340_0,
        &reshape_338_0_dim_0,
        &reshape_340_0_dim_2,
        &reshape_338_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_341" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_343" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_341_0,
        encoder_layers_22_mha_proj_weight,
        encoder_layers_22_mha_proj_bias,
        gemm_rcr_bias_add_332_0,

        reshape_344_0,
        global_workspace_,

     1,


        &reshape_342_0_dim_0,

        &reshape_342_0_dim_1,


        &encoder_layers_22_mha_proj_weight_dim_0,

        &encoder_layers_22_mha_proj_weight_dim_1,


        &reshape_342_0_dim_0,

        &encoder_layers_22_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_343" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_345" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_344_0_dim_0;

        M *= reshape_344_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_344_0_dim_2;

    
        layernorm_15(
           layernorm_345_0, reshape_344_0, encoder_layers_22_ln2_weight, encoder_layers_22_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_345" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_346" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_345_0,
        encoder_layers_22_mlp_fc1_weight,

        encoder_layers_22_mlp_fc1_bias,

        gemm_rcr_bias_gelu_346_0,
        global_workspace_,
        1,

        &reshape_344_0_dim_0,

        &reshape_344_0_dim_1,

        &reshape_344_0_dim_2,


        &encoder_layers_22_mlp_fc1_weight_dim_0,

        &encoder_layers_22_mlp_fc1_weight_dim_1,


        &reshape_344_0_dim_0,

        &reshape_344_0_dim_1,

        &encoder_layers_22_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_346" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_347" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_346_0,
        encoder_layers_22_mlp_fc2_weight,
        encoder_layers_22_mlp_fc2_bias,
        reshape_344_0,

        gemm_rcr_bias_add_347_0,
        global_workspace_,

     1,


        &reshape_344_0_dim_0,

        &reshape_344_0_dim_1,

        &encoder_layers_22_mlp_fc1_weight_dim_0,


        &encoder_layers_22_mlp_fc2_weight_dim_0,

        &encoder_layers_22_mlp_fc2_weight_dim_1,


        &reshape_344_0_dim_0,

        &reshape_344_0_dim_1,

        &encoder_layers_22_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_347" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_348" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_344_0_dim_0;

        M *= reshape_344_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_22_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_349_0, gemm_rcr_bias_add_347_0, encoder_layers_23_ln1_weight, encoder_layers_23_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_348" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_350" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_349_0,
        encoder_layers_23_mha_qkv_weight,

        encoder_layers_23_mha_qkv_bias,

        reshape_351_0,
        global_workspace_,
        1,

        &reshape_349_0_dim_0,

        &reshape_349_0_dim_1,


        &encoder_layers_23_mha_qkv_weight_dim_0,

        &encoder_layers_23_mha_qkv_weight_dim_1,


        &reshape_349_0_dim_0,

        &encoder_layers_23_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_350" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_352" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_352_1,
      split_352_2,
      split_352_0
    };


      int64_t *split_352_1_shape[] = {
        &split_352_1_dim_0, &reshape_351_0_dim_1, &reshape_351_0_dim_2, &reshape_351_0_dim_3, &reshape_351_0_dim_4
      };
      int64_t *split_352_2_shape[] = {
        &split_352_2_dim_0, &reshape_351_0_dim_1, &reshape_351_0_dim_2, &reshape_351_0_dim_3, &reshape_351_0_dim_4
      };
      int64_t *split_352_0_shape[] = {
        &split_352_0_dim_0, &reshape_351_0_dim_1, &reshape_351_0_dim_2, &reshape_351_0_dim_3, &reshape_351_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_352_1_shape, split_352_2_shape, split_352_0_shape
    };

    const int64_t reshape_351_0_shape[] = {
      reshape_351_0_dim_0, reshape_351_0_dim_1, reshape_351_0_dim_2, reshape_351_0_dim_3, reshape_351_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_351_0,
        reshape_351_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_352" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_353" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_352_0_dim_0,
        &reshape_351_0_dim_1,
        &reshape_351_0_dim_2,
        &reshape_351_0_dim_3,
        &reshape_351_0_dim_4,
        &reshape_353_0_dim_0,
        &reshape_353_0_dim_1,
        &reshape_353_0_dim_2,
        &reshape_353_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_353" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_354" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_352_1_dim_0,
        &reshape_351_0_dim_1,
        &reshape_351_0_dim_2,
        &reshape_351_0_dim_3,
        &reshape_351_0_dim_4,
        &reshape_354_0_dim_0,
        &reshape_354_0_dim_1,
        &reshape_354_0_dim_2,
        &reshape_354_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_354" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_355" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_352_2_dim_0,
        &reshape_351_0_dim_1,
        &reshape_351_0_dim_2,
        &reshape_351_0_dim_3,
        &reshape_351_0_dim_4,
        &reshape_355_0_dim_0,
        &reshape_355_0_dim_1,
        &reshape_355_0_dim_2,
        &reshape_355_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_355" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_356" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_356_0,
        reshape_354_0, reshape_355_0, reshape_353_0,
        &reshape_354_0_dim_0,
        &reshape_353_0_dim_2,
        &reshape_354_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_356" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_358" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_356_0,
        encoder_layers_23_mha_proj_weight,
        encoder_layers_23_mha_proj_bias,
        gemm_rcr_bias_add_347_0,

        reshape_359_0,
        global_workspace_,

     1,


        &reshape_357_0_dim_0,

        &reshape_357_0_dim_1,


        &encoder_layers_23_mha_proj_weight_dim_0,

        &encoder_layers_23_mha_proj_weight_dim_1,


        &reshape_357_0_dim_0,

        &encoder_layers_23_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_358" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_360" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_359_0_dim_0;

        M *= reshape_359_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_359_0_dim_2;

    
        layernorm_15(
           layernorm_360_0, reshape_359_0, encoder_layers_23_ln2_weight, encoder_layers_23_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_360" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_361" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_360_0,
        encoder_layers_23_mlp_fc1_weight,

        encoder_layers_23_mlp_fc1_bias,

        gemm_rcr_bias_gelu_361_0,
        global_workspace_,
        1,

        &reshape_359_0_dim_0,

        &reshape_359_0_dim_1,

        &reshape_359_0_dim_2,


        &encoder_layers_23_mlp_fc1_weight_dim_0,

        &encoder_layers_23_mlp_fc1_weight_dim_1,


        &reshape_359_0_dim_0,

        &reshape_359_0_dim_1,

        &encoder_layers_23_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_361" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_362" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_361_0,
        encoder_layers_23_mlp_fc2_weight,
        encoder_layers_23_mlp_fc2_bias,
        reshape_359_0,

        gemm_rcr_bias_add_362_0,
        global_workspace_,

     1,


        &reshape_359_0_dim_0,

        &reshape_359_0_dim_1,

        &encoder_layers_23_mlp_fc1_weight_dim_0,


        &encoder_layers_23_mlp_fc2_weight_dim_0,

        &encoder_layers_23_mlp_fc2_weight_dim_1,


        &reshape_359_0_dim_0,

        &reshape_359_0_dim_1,

        &encoder_layers_23_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_362" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_363" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_359_0_dim_0;

        M *= reshape_359_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_23_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_364_0, gemm_rcr_bias_add_362_0, encoder_layers_24_ln1_weight, encoder_layers_24_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_363" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_365" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_364_0,
        encoder_layers_24_mha_qkv_weight,

        encoder_layers_24_mha_qkv_bias,

        reshape_366_0,
        global_workspace_,
        1,

        &reshape_364_0_dim_0,

        &reshape_364_0_dim_1,


        &encoder_layers_24_mha_qkv_weight_dim_0,

        &encoder_layers_24_mha_qkv_weight_dim_1,


        &reshape_364_0_dim_0,

        &encoder_layers_24_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_365" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_367" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_367_0,
      split_367_2,
      split_367_1
    };


      int64_t *split_367_0_shape[] = {
        &split_367_0_dim_0, &reshape_366_0_dim_1, &reshape_366_0_dim_2, &reshape_366_0_dim_3, &reshape_366_0_dim_4
      };
      int64_t *split_367_2_shape[] = {
        &split_367_2_dim_0, &reshape_366_0_dim_1, &reshape_366_0_dim_2, &reshape_366_0_dim_3, &reshape_366_0_dim_4
      };
      int64_t *split_367_1_shape[] = {
        &split_367_1_dim_0, &reshape_366_0_dim_1, &reshape_366_0_dim_2, &reshape_366_0_dim_3, &reshape_366_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_367_0_shape, split_367_2_shape, split_367_1_shape
    };

    const int64_t reshape_366_0_shape[] = {
      reshape_366_0_dim_0, reshape_366_0_dim_1, reshape_366_0_dim_2, reshape_366_0_dim_3, reshape_366_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_366_0,
        reshape_366_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_367" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_368" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_367_0_dim_0,
        &reshape_366_0_dim_1,
        &reshape_366_0_dim_2,
        &reshape_366_0_dim_3,
        &reshape_366_0_dim_4,
        &reshape_368_0_dim_0,
        &reshape_368_0_dim_1,
        &reshape_368_0_dim_2,
        &reshape_368_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_368" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_369" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_367_1_dim_0,
        &reshape_366_0_dim_1,
        &reshape_366_0_dim_2,
        &reshape_366_0_dim_3,
        &reshape_366_0_dim_4,
        &reshape_369_0_dim_0,
        &reshape_369_0_dim_1,
        &reshape_369_0_dim_2,
        &reshape_369_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_369" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_370" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_367_2_dim_0,
        &reshape_366_0_dim_1,
        &reshape_366_0_dim_2,
        &reshape_366_0_dim_3,
        &reshape_366_0_dim_4,
        &reshape_370_0_dim_0,
        &reshape_370_0_dim_1,
        &reshape_370_0_dim_2,
        &reshape_370_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_370" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_371" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_371_0,
        reshape_368_0, reshape_370_0, reshape_369_0,
        &reshape_368_0_dim_0,
        &reshape_369_0_dim_2,
        &reshape_368_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_371" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_373" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_371_0,
        encoder_layers_24_mha_proj_weight,
        encoder_layers_24_mha_proj_bias,
        gemm_rcr_bias_add_362_0,

        reshape_374_0,
        global_workspace_,

     1,


        &reshape_372_0_dim_0,

        &reshape_372_0_dim_1,


        &encoder_layers_24_mha_proj_weight_dim_0,

        &encoder_layers_24_mha_proj_weight_dim_1,


        &reshape_372_0_dim_0,

        &encoder_layers_24_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_373" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_375" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_374_0_dim_0;

        M *= reshape_374_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_374_0_dim_2;

    
        layernorm_15(
           layernorm_375_0, reshape_374_0, encoder_layers_24_ln2_weight, encoder_layers_24_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_375" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_376" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_375_0,
        encoder_layers_24_mlp_fc1_weight,

        encoder_layers_24_mlp_fc1_bias,

        gemm_rcr_bias_gelu_376_0,
        global_workspace_,
        1,

        &reshape_374_0_dim_0,

        &reshape_374_0_dim_1,

        &reshape_374_0_dim_2,


        &encoder_layers_24_mlp_fc1_weight_dim_0,

        &encoder_layers_24_mlp_fc1_weight_dim_1,


        &reshape_374_0_dim_0,

        &reshape_374_0_dim_1,

        &encoder_layers_24_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_376" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_377" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_376_0,
        encoder_layers_24_mlp_fc2_weight,
        encoder_layers_24_mlp_fc2_bias,
        reshape_374_0,

        gemm_rcr_bias_add_377_0,
        global_workspace_,

     1,


        &reshape_374_0_dim_0,

        &reshape_374_0_dim_1,

        &encoder_layers_24_mlp_fc1_weight_dim_0,


        &encoder_layers_24_mlp_fc2_weight_dim_0,

        &encoder_layers_24_mlp_fc2_weight_dim_1,


        &reshape_374_0_dim_0,

        &reshape_374_0_dim_1,

        &encoder_layers_24_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_377" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_378" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_374_0_dim_0;

        M *= reshape_374_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_24_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_379_0, gemm_rcr_bias_add_377_0, encoder_layers_25_ln1_weight, encoder_layers_25_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_378" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_380" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_379_0,
        encoder_layers_25_mha_qkv_weight,

        encoder_layers_25_mha_qkv_bias,

        reshape_381_0,
        global_workspace_,
        1,

        &reshape_379_0_dim_0,

        &reshape_379_0_dim_1,


        &encoder_layers_25_mha_qkv_weight_dim_0,

        &encoder_layers_25_mha_qkv_weight_dim_1,


        &reshape_379_0_dim_0,

        &encoder_layers_25_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_380" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_382" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_382_1,
      split_382_0,
      split_382_2
    };


      int64_t *split_382_1_shape[] = {
        &split_382_1_dim_0, &reshape_381_0_dim_1, &reshape_381_0_dim_2, &reshape_381_0_dim_3, &reshape_381_0_dim_4
      };
      int64_t *split_382_0_shape[] = {
        &split_382_0_dim_0, &reshape_381_0_dim_1, &reshape_381_0_dim_2, &reshape_381_0_dim_3, &reshape_381_0_dim_4
      };
      int64_t *split_382_2_shape[] = {
        &split_382_2_dim_0, &reshape_381_0_dim_1, &reshape_381_0_dim_2, &reshape_381_0_dim_3, &reshape_381_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_382_1_shape, split_382_0_shape, split_382_2_shape
    };

    const int64_t reshape_381_0_shape[] = {
      reshape_381_0_dim_0, reshape_381_0_dim_1, reshape_381_0_dim_2, reshape_381_0_dim_3, reshape_381_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_381_0,
        reshape_381_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_382" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_383" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_382_0_dim_0,
        &reshape_381_0_dim_1,
        &reshape_381_0_dim_2,
        &reshape_381_0_dim_3,
        &reshape_381_0_dim_4,
        &reshape_383_0_dim_0,
        &reshape_383_0_dim_1,
        &reshape_383_0_dim_2,
        &reshape_383_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_383" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_384" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_382_1_dim_0,
        &reshape_381_0_dim_1,
        &reshape_381_0_dim_2,
        &reshape_381_0_dim_3,
        &reshape_381_0_dim_4,
        &reshape_384_0_dim_0,
        &reshape_384_0_dim_1,
        &reshape_384_0_dim_2,
        &reshape_384_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_384" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_385" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_382_2_dim_0,
        &reshape_381_0_dim_1,
        &reshape_381_0_dim_2,
        &reshape_381_0_dim_3,
        &reshape_381_0_dim_4,
        &reshape_385_0_dim_0,
        &reshape_385_0_dim_1,
        &reshape_385_0_dim_2,
        &reshape_385_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_385" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_386" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_386_0,
        reshape_384_0, reshape_383_0, reshape_385_0,
        &reshape_384_0_dim_0,
        &reshape_385_0_dim_2,
        &reshape_384_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_386" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_388" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_386_0,
        encoder_layers_25_mha_proj_weight,
        encoder_layers_25_mha_proj_bias,
        gemm_rcr_bias_add_377_0,

        reshape_389_0,
        global_workspace_,

     1,


        &reshape_387_0_dim_0,

        &reshape_387_0_dim_1,


        &encoder_layers_25_mha_proj_weight_dim_0,

        &encoder_layers_25_mha_proj_weight_dim_1,


        &reshape_387_0_dim_0,

        &encoder_layers_25_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_388" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_390" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_389_0_dim_0;

        M *= reshape_389_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_389_0_dim_2;

    
        layernorm_15(
           layernorm_390_0, reshape_389_0, encoder_layers_25_ln2_weight, encoder_layers_25_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_390" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_391" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_390_0,
        encoder_layers_25_mlp_fc1_weight,

        encoder_layers_25_mlp_fc1_bias,

        gemm_rcr_bias_gelu_391_0,
        global_workspace_,
        1,

        &reshape_389_0_dim_0,

        &reshape_389_0_dim_1,

        &reshape_389_0_dim_2,


        &encoder_layers_25_mlp_fc1_weight_dim_0,

        &encoder_layers_25_mlp_fc1_weight_dim_1,


        &reshape_389_0_dim_0,

        &reshape_389_0_dim_1,

        &encoder_layers_25_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_391" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_392" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_391_0,
        encoder_layers_25_mlp_fc2_weight,
        encoder_layers_25_mlp_fc2_bias,
        reshape_389_0,

        gemm_rcr_bias_add_392_0,
        global_workspace_,

     1,


        &reshape_389_0_dim_0,

        &reshape_389_0_dim_1,

        &encoder_layers_25_mlp_fc1_weight_dim_0,


        &encoder_layers_25_mlp_fc2_weight_dim_0,

        &encoder_layers_25_mlp_fc2_weight_dim_1,


        &reshape_389_0_dim_0,

        &reshape_389_0_dim_1,

        &encoder_layers_25_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_392" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_393" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_389_0_dim_0;

        M *= reshape_389_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_25_mlp_fc2_weight_dim_0;

    
        layernorm_3(
           reshape_394_0, gemm_rcr_bias_add_392_0, encoder_layers_26_ln1_weight, encoder_layers_26_ln1_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_393" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_permute_395" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_permute_5(

        reshape_394_0,
        encoder_layers_26_mha_qkv_weight,

        encoder_layers_26_mha_qkv_bias,

        reshape_396_0,
        global_workspace_,
        1,

        &reshape_394_0_dim_0,

        &reshape_394_0_dim_1,


        &encoder_layers_26_mha_qkv_weight_dim_0,

        &encoder_layers_26_mha_qkv_weight_dim_1,


        &reshape_394_0_dim_0,

        &encoder_layers_26_mha_qkv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_permute_395" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"3456\", \"1152\"], [\"3456\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"3456\"]]"
        
          << ", \"layout\": " << "\"20314\""
        
          << ", \"shape\": " << "\"(729, 3, 16)\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_397" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_397_1,
      split_397_2,
      split_397_0
    };


      int64_t *split_397_1_shape[] = {
        &split_397_1_dim_0, &reshape_396_0_dim_1, &reshape_396_0_dim_2, &reshape_396_0_dim_3, &reshape_396_0_dim_4
      };
      int64_t *split_397_2_shape[] = {
        &split_397_2_dim_0, &reshape_396_0_dim_1, &reshape_396_0_dim_2, &reshape_396_0_dim_3, &reshape_396_0_dim_4
      };
      int64_t *split_397_0_shape[] = {
        &split_397_0_dim_0, &reshape_396_0_dim_1, &reshape_396_0_dim_2, &reshape_396_0_dim_3, &reshape_396_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_397_1_shape, split_397_2_shape, split_397_0_shape
    };

    const int64_t reshape_396_0_shape[] = {
      reshape_396_0_dim_0, reshape_396_0_dim_1, reshape_396_0_dim_2, reshape_396_0_dim_3, reshape_396_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1, 1
    };

    bool output_masks[] = {
      true, true, true
    };

    split_7(
        outputs,
        output_shapes,
        output_masks,
        reshape_396_0,
        reshape_396_0_shape,
        3/*real_num_splits*/,
        3/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_397" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"3\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_398" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_397_0_dim_0,
        &reshape_396_0_dim_1,
        &reshape_396_0_dim_2,
        &reshape_396_0_dim_3,
        &reshape_396_0_dim_4,
        &reshape_398_0_dim_0,
        &reshape_398_0_dim_1,
        &reshape_398_0_dim_2,
        &reshape_398_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_398" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_399" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_397_1_dim_0,
        &reshape_396_0_dim_1,
        &reshape_396_0_dim_2,
        &reshape_396_0_dim_3,
        &reshape_396_0_dim_4,
        &reshape_399_0_dim_0,
        &reshape_399_0_dim_1,
        &reshape_399_0_dim_2,
        &reshape_399_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_399" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_400" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_8(
        &split_397_2_dim_0,
        &reshape_396_0_dim_1,
        &reshape_396_0_dim_2,
        &reshape_396_0_dim_3,
        &reshape_396_0_dim_4,
        &reshape_400_0_dim_0,
        &reshape_400_0_dim_1,
        &reshape_400_0_dim_2,
        &reshape_400_0_dim_3
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_400" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_401" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_11(
        mem_eff_attention_401_0,
        reshape_399_0, reshape_400_0, reshape_398_0,
        &reshape_399_0_dim_0,
        &reshape_398_0_dim_2,
        &reshape_399_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_401" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_403" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_13(

        mem_eff_attention_401_0,
        encoder_layers_26_mha_proj_weight,
        encoder_layers_26_mha_proj_bias,
        gemm_rcr_bias_add_392_0,

        reshape_404_0,
        global_workspace_,

     1,


        &reshape_402_0_dim_0,

        &reshape_402_0_dim_1,


        &encoder_layers_26_mha_proj_weight_dim_0,

        &encoder_layers_26_mha_proj_weight_dim_1,


        &reshape_402_0_dim_0,

        &encoder_layers_26_mha_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_403" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"5832\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"5832\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_405" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_404_0_dim_0;

        M *= reshape_404_0_dim_1;

    

        int64_t N = 1;

        N *= reshape_404_0_dim_2;

    
        layernorm_15(
           layernorm_405_0, reshape_404_0, encoder_layers_26_ln2_weight, encoder_layers_26_ln2_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_405" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_406" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_16(

        layernorm_405_0,
        encoder_layers_26_mlp_fc1_weight,

        encoder_layers_26_mlp_fc1_bias,

        gemm_rcr_bias_gelu_406_0,
        global_workspace_,
        1,

        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &reshape_404_0_dim_2,


        &encoder_layers_26_mlp_fc1_weight_dim_0,

        &encoder_layers_26_mlp_fc1_weight_dim_1,


        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &encoder_layers_26_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_406" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_407" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_17(

        gemm_rcr_bias_gelu_406_0,
        encoder_layers_26_mlp_fc2_weight,
        encoder_layers_26_mlp_fc2_bias,
        reshape_404_0,

        gemm_rcr_bias_add_407_0,
        global_workspace_,

     1,


        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &encoder_layers_26_mlp_fc1_weight_dim_0,


        &encoder_layers_26_mlp_fc2_weight_dim_0,

        &encoder_layers_26_mlp_fc2_weight_dim_1,


        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &encoder_layers_26_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_407" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_408" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_404_0_dim_0;

        M *= reshape_404_0_dim_1;

    

        int64_t N = 1;

        N *= encoder_layers_26_mlp_fc2_weight_dim_0;

    
        layernorm_15(
           layernorm_408_0, gemm_rcr_bias_add_407_0, encoder_ln_weight, encoder_ln_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_408" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_409" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_409(

        layernorm_408_0,
        pool_kv_weight,

        pool_kv_bias,

        reshape_410_0,
        global_workspace_,
        1,

        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &encoder_layers_26_mlp_fc2_weight_dim_0,


        &pool_kv_weight_dim_0,

        &pool_kv_weight_dim_1,


        &reshape_404_0_dim_0,

        &reshape_404_0_dim_1,

        &pool_kv_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_409" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"1152\"], [\"2304\", \"1152\"], [\"2304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"729\", \"2304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "permute_411" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
{
    int dims[] = {2,0,3,1,4};
    permute_411(
        permute_411_0,
        reshape_410_0,

        &reshape_410_0_dim_0,

        &reshape_410_0_dim_1,

        &reshape_410_0_dim_2,

        &reshape_410_0_dim_3,

        &reshape_410_0_dim_4,

        dims,
        stream
    );
}
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "permute_411" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"729\", \"2\", \"16\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"2\", \"8\", \"16\", \"729\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "split_412" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
  {

    void *outputs[] = {
      split_412_1,
      split_412_0
    };


      int64_t *split_412_1_shape[] = {
        &split_412_1_dim_0, &reshape_410_0_dim_0, &reshape_410_0_dim_3, &reshape_410_0_dim_1, &reshape_410_0_dim_4
      };
      int64_t *split_412_0_shape[] = {
        &split_412_0_dim_0, &reshape_410_0_dim_0, &reshape_410_0_dim_3, &reshape_410_0_dim_1, &reshape_410_0_dim_4
      };

    int64_t **output_shapes[] = {
      split_412_1_shape, split_412_0_shape
    };

    const int64_t permute_411_0_shape[] = {
      reshape_410_0_dim_2, reshape_410_0_dim_0, reshape_410_0_dim_3, reshape_410_0_dim_1, reshape_410_0_dim_4
    };

    int64_t split_sizes[] = {
      1, 1
    };

    bool output_masks[] = {
      true, true
    };

    split_412(
        outputs,
        output_shapes,
        output_masks,
        permute_411_0,
        permute_411_0_shape,
        2/*real_num_splits*/,
        2/*all_num_splits*/,
        split_sizes,
        0/*split_dim*/,
        5/*rank*/,
        stream
    );
  }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "split_412" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"2\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"], [\"1\", \"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"split_sizes\": " << "\"[1, 1]]\""
        
          << ", \"dim\": " << "\"0]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "squeeze_413" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_squeeze_413(
        &split_412_0_dim_0,
        &reshape_410_0_dim_0,
        &reshape_410_0_dim_3,
        &reshape_410_0_dim_1,
        &reshape_410_0_dim_4,
        &reshape_410_0_dim_0,
        &reshape_410_0_dim_3,
        &reshape_410_0_dim_1,
        &reshape_410_0_dim_4
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "squeeze_413" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"dim\": " << "\"0\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "squeeze_414" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_squeeze_413(
        &split_412_1_dim_0,
        &reshape_410_0_dim_0,
        &reshape_410_0_dim_3,
        &reshape_410_0_dim_1,
        &reshape_410_0_dim_4,
        &reshape_410_0_dim_0,
        &reshape_410_0_dim_3,
        &reshape_410_0_dim_1,
        &reshape_410_0_dim_4
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "squeeze_414" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"1\", \"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"729\", \"72\"]]"
        
          << ", \"dim\": " << "\"0\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "mem_eff_attention_418" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    mem_eff_attention_418(
        mem_eff_attention_418_0,
        reshape_417_0, squeeze_414_0, squeeze_413_0,
        &reshape_417_0_dim_0,
        &reshape_410_0_dim_1,
        &reshape_417_0_dim_2,
        16,
        72,
        72,
        0,
        0.11785113019775792,
        false,
        true,
        static_cast<int32_t*>(nullptr),
        true,
        static_cast<int32_t*>(nullptr),
        global_workspace_,
        stream
    );
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "mem_eff_attention_418" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"16\", \"1\", \"72\"], [\"8\", \"16\", \"729\", \"72\"], [\"8\", \"16\", \"729\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"1\", \"16\", \"72\"]]"
        
          << ", \"causal\": " << "\"False\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "reshape_426" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    ait_reshape_426(
        &reshape_417_0_dim_0,
        &reshape_417_0_dim_2,
        &reshape_417_0_dim_1,
        &reshape_410_0_dim_4,
        &reshape_417_0_dim_0,
        &reshape_417_0_dim_1,
        &reshape_417_0_dim_2,
        &reshape_410_0_dim_4
    );
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "reshape_426" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"1\", \"16\", \"72\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"16\", \"1\", \"72\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_421" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_421(

        reshape_426_0,
        pool_proj_weight,

        pool_proj_bias,

        gemm_rcr_bias_421_0,
        global_workspace_,
        1,

        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &reshape_420_0_dim_2,


        &pool_proj_weight_dim_0,

        &pool_proj_weight_dim_1,


        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &pool_proj_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_421" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"1\", \"1152\"], [\"1152\", \"1152\"], [\"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"1\", \"1152\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "layernorm_422" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
      {
        
        int64_t M = 1;

        M *= reshape_420_0_dim_0;

        M *= reshape_420_0_dim_1;

    

        int64_t N = 1;

        N *= pool_proj_weight_dim_0;

    
        layernorm_422(
           layernorm_422_0, gemm_rcr_bias_421_0, pool_ln_weight, pool_ln_bias,
           M, N, 1e-05, stream /* default stream */
        );
      }
    
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "layernorm_422" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"1\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"1\", \"1152\"]]"
        
          << ", \"normalized_shape\": " << "\"[1152]\""
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_gelu_423" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_gelu_423(

        layernorm_422_0,
        pool_mlp_fc1_weight,

        pool_mlp_fc1_bias,

        gemm_rcr_bias_gelu_423_0,
        global_workspace_,
        1,

        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &pool_proj_weight_dim_0,


        &pool_mlp_fc1_weight_dim_0,

        &pool_mlp_fc1_weight_dim_1,


        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &pool_mlp_fc1_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_gelu_423" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"1\", \"1152\"], [\"4304\", \"1152\"], [\"4304\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"1\", \"4304\"]]"
        
           << " } ";
        
          ss << ",\n";
        
      }
      
      {
        std::cout << "Profiling: " << "gemm_rcr_bias_add_424" << " (" << iters << " iterations)" << std::endl;
        std::vector<std::pair<EventType, EventType>> call_events(iters);
        for (auto& [call_start, call_end] : call_events) {
          CreateEvent(&call_start);
          CreateEvent(&call_end);
        }
        for (auto& [call_start, call_end]: call_events) {
          DeviceMemset(L2CacheSlab, 0x73, L2SizeInBytes);
          EventRecord(call_start, stream);
            
    {
    

    gemm_rcr_bias_add_424(

        gemm_rcr_bias_gelu_423_0,
        pool_mlp_fc2_weight,
        pool_mlp_fc2_bias,
        gemm_rcr_bias_421_0,

        output_0,
        global_workspace_,

     2,


        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &pool_mlp_fc1_weight_dim_0,


        &pool_mlp_fc2_weight_dim_0,

        &pool_mlp_fc2_weight_dim_1,


        &reshape_420_0_dim_0,

        &reshape_420_0_dim_1,

        &pool_mlp_fc2_weight_dim_0,

        stream
    );
    }
          EventRecord(call_end, stream);
          DeviceCheckLastError(__FILE__, __LINE__);
        }
        EventSynchronize(std::get<1>(call_events.back()));
        float milliseconds = 0.0;
        for (auto& [call_start, call_end] : call_events) {
          float call_milliseconds = 0.0;
          EventElapsedTime(&call_milliseconds, call_start, call_end);
          DestroyEvent(call_start);
          DestroyEvent(call_end);
          milliseconds += call_milliseconds;
        }
        ss << "\"" << "gemm_rcr_bias_add_424" << "\": { \"ms_per_iter\": "
           << std::setprecision(4) << (milliseconds/iters)
           << ", \"qps\": " << 1000 * iters / milliseconds
           << ", \"input_sizes\": " << "[[\"8\", \"1\", \"4304\"], [\"1152\", \"4304\"], [\"1152\"], [\"8\", \"1\", \"1152\"]]"
           << ", \"output_sizes\": " << "[[\"8\", \"1\", \"1152\"]]"
        
           << " } ";
        
          ss << "\n";
        
      }
      
      ss << "}\n";

      DeviceToDeviceCopies(stream);
      std::cout << "AIT per op profiling finished." << std::endl;
      FreeDeviceMemory(L2CacheSlab);
#endif
    }

    static std::unique_ptr<Model> Create(
      AITemplateAllocator& allocator,
      uint8_t* constants
    ) {
      return std::make_unique<Model>(
          103949568,
          26873856 * (1 + 0),
          0 * (1 + 0),
          1,
          1,
          342,
          constants,
          allocator
      );
    }

  private:
   void* input0 {nullptr};
   void* patch_embed_proj_weight {nullptr};
   void* patch_embed_proj_bias {nullptr};
   void* pos_emb_pos_emb {nullptr};
   void* encoder_layers_0_ln1_weight {nullptr};
   void* encoder_layers_0_ln1_bias {nullptr};
   void* encoder_layers_0_mha_qkv_weight {nullptr};
   void* encoder_layers_0_mha_qkv_bias {nullptr};
   void* encoder_layers_0_mha_proj_weight {nullptr};
   void* encoder_layers_0_mha_proj_bias {nullptr};
   void* encoder_layers_0_ln2_weight {nullptr};
   void* encoder_layers_0_ln2_bias {nullptr};
   void* encoder_layers_0_mlp_fc1_weight {nullptr};
   void* encoder_layers_0_mlp_fc1_bias {nullptr};
   void* encoder_layers_0_mlp_fc2_weight {nullptr};
   void* encoder_layers_0_mlp_fc2_bias {nullptr};
   void* encoder_layers_1_ln1_weight {nullptr};
   void* encoder_layers_1_ln1_bias {nullptr};
   void* encoder_layers_1_mha_qkv_weight {nullptr};
   void* encoder_layers_1_mha_qkv_bias {nullptr};
   void* encoder_layers_1_mha_proj_weight {nullptr};
   void* encoder_layers_1_mha_proj_bias {nullptr};
   void* encoder_layers_1_ln2_weight {nullptr};
   void* encoder_layers_1_ln2_bias {nullptr};
   void* encoder_layers_1_mlp_fc1_weight {nullptr};
   void* encoder_layers_1_mlp_fc1_bias {nullptr};
   void* encoder_layers_1_mlp_fc2_weight {nullptr};
   void* encoder_layers_1_mlp_fc2_bias {nullptr};
   void* encoder_layers_2_ln1_weight {nullptr};
   void* encoder_layers_2_ln1_bias {nullptr};
   void* encoder_layers_2_mha_qkv_weight {nullptr};
   void* encoder_layers_2_mha_qkv_bias {nullptr};
   void* encoder_layers_2_mha_proj_weight {nullptr};
   void* encoder_layers_2_mha_proj_bias {nullptr};
   void* encoder_layers_2_ln2_weight {nullptr};
   void* encoder_layers_2_ln2_bias {nullptr};
   void* encoder_layers_2_mlp_fc1_weight {nullptr};
   void* encoder_layers_2_mlp_fc1_bias {nullptr};
   void* encoder_layers_2_mlp_fc2_weight {nullptr};
   void* encoder_layers_2_mlp_fc2_bias {nullptr};
   void* encoder_layers_3_ln1_weight {nullptr};
   void* encoder_layers_3_ln1_bias {nullptr};
   void* encoder_layers_3_mha_qkv_weight {nullptr};
   void* encoder_layers_3_mha_qkv_bias {nullptr};
   void* encoder_layers_3_mha_proj_weight {nullptr};
   void* encoder_layers_3_mha_proj_bias {nullptr};
   void* encoder_layers_3_ln2_weight {nullptr};
   void* encoder_layers_3_ln2_bias {nullptr};
   void* encoder_layers_3_mlp_fc1_weight {nullptr};
   void* encoder_layers_3_mlp_fc1_bias {nullptr};
   void* encoder_layers_3_mlp_fc2_weight {nullptr};
   void* encoder_layers_3_mlp_fc2_bias {nullptr};
   void* encoder_layers_4_ln1_weight {nullptr};
   void* encoder_layers_4_ln1_bias {nullptr};
   void* encoder_layers_4_mha_qkv_weight {nullptr};
   void* encoder_layers_4_mha_qkv_bias {nullptr};
   void* encoder_layers_4_mha_proj_weight {nullptr};
   void* encoder_layers_4_mha_proj_bias {nullptr};
   void* encoder_layers_4_ln2_weight {nullptr};
   void* encoder_layers_4_ln2_bias {nullptr};
   void* encoder_layers_4_mlp_fc1_weight {nullptr};
   void* encoder_layers_4_mlp_fc1_bias {nullptr};
   void* encoder_layers_4_mlp_fc2_weight {nullptr};
   void* encoder_layers_4_mlp_fc2_bias {nullptr};
   void* encoder_layers_5_ln1_weight {nullptr};
   void* encoder_layers_5_ln1_bias {nullptr};
   void* encoder_layers_5_mha_qkv_weight {nullptr};
   void* encoder_layers_5_mha_qkv_bias {nullptr};
   void* encoder_layers_5_mha_proj_weight {nullptr};
   void* encoder_layers_5_mha_proj_bias {nullptr};
   void* encoder_layers_5_ln2_weight {nullptr};
   void* encoder_layers_5_ln2_bias {nullptr};
   void* encoder_layers_5_mlp_fc1_weight {nullptr};
   void* encoder_layers_5_mlp_fc1_bias {nullptr};
   void* encoder_layers_5_mlp_fc2_weight {nullptr};
   void* encoder_layers_5_mlp_fc2_bias {nullptr};
   void* encoder_layers_6_ln1_weight {nullptr};
   void* encoder_layers_6_ln1_bias {nullptr};
   void* encoder_layers_6_mha_qkv_weight {nullptr};
   void* encoder_layers_6_mha_qkv_bias {nullptr};
   void* encoder_layers_6_mha_proj_weight {nullptr};
   void* encoder_layers_6_mha_proj_bias {nullptr};
   void* encoder_layers_6_ln2_weight {nullptr};
   void* encoder_layers_6_ln2_bias {nullptr};
   void* encoder_layers_6_mlp_fc1_weight {nullptr};
   void* encoder_layers_6_mlp_fc1_bias {nullptr};
   void* encoder_layers_6_mlp_fc2_weight {nullptr};
   void* encoder_layers_6_mlp_fc2_bias {nullptr};
   void* encoder_layers_7_ln1_weight {nullptr};
   void* encoder_layers_7_ln1_bias {nullptr};
   void* encoder_layers_7_mha_qkv_weight {nullptr};
   void* encoder_layers_7_mha_qkv_bias {nullptr};
   void* encoder_layers_7_mha_proj_weight {nullptr};
   void* encoder_layers_7_mha_proj_bias {nullptr};
   void* encoder_layers_7_ln2_weight {nullptr};
   void* encoder_layers_7_ln2_bias {nullptr};
   void* encoder_layers_7_mlp_fc1_weight {nullptr};
   void* encoder_layers_7_mlp_fc1_bias {nullptr};
   void* encoder_layers_7_mlp_fc2_weight {nullptr};
   void* encoder_layers_7_mlp_fc2_bias {nullptr};
   void* encoder_layers_8_ln1_weight {nullptr};
   void* encoder_layers_8_ln1_bias {nullptr};
   void* encoder_layers_8_mha_qkv_weight {nullptr};
   void* encoder_layers_8_mha_qkv_bias {nullptr};
   void* encoder_layers_8_mha_proj_weight {nullptr};
   void* encoder_layers_8_mha_proj_bias {nullptr};
   void* encoder_layers_8_ln2_weight {nullptr};
   void* encoder_layers_8_ln2_bias {nullptr};
   void* encoder_layers_8_mlp_fc1_weight {nullptr};
   void* encoder_layers_8_mlp_fc1_bias {nullptr};
   void* encoder_layers_8_mlp_fc2_weight {nullptr};
   void* encoder_layers_8_mlp_fc2_bias {nullptr};
   void* encoder_layers_9_ln1_weight {nullptr};
   void* encoder_layers_9_ln1_bias {nullptr};
   void* encoder_layers_9_mha_qkv_weight {nullptr};
   void* encoder_layers_9_mha_qkv_bias {nullptr};
   void* encoder_layers_9_mha_proj_weight {nullptr};
   void* encoder_layers_9_mha_proj_bias {nullptr};
   void* encoder_layers_9_ln2_weight {nullptr};
   void* encoder_layers_9_ln2_bias {nullptr};
   void* encoder_layers_9_mlp_fc1_weight {nullptr};
   void* encoder_layers_9_mlp_fc1_bias {nullptr};
   void* encoder_layers_9_mlp_fc2_weight {nullptr};
   void* encoder_layers_9_mlp_fc2_bias {nullptr};
   void* encoder_layers_10_ln1_weight {nullptr};
   void* encoder_layers_10_ln1_bias {nullptr};
   void* encoder_layers_10_mha_qkv_weight {nullptr};
   void* encoder_layers_10_mha_qkv_bias {nullptr};
   void* encoder_layers_10_mha_proj_weight {nullptr};
   void* encoder_layers_10_mha_proj_bias {nullptr};
   void* encoder_layers_10_ln2_weight {nullptr};
   void* encoder_layers_10_ln2_bias {nullptr};
   void* encoder_layers_10_mlp_fc1_weight {nullptr};
   void* encoder_layers_10_mlp_fc1_bias {nullptr};
   void* encoder_layers_10_mlp_fc2_weight {nullptr};
   void* encoder_layers_10_mlp_fc2_bias {nullptr};
   void* encoder_layers_11_ln1_weight {nullptr};
   void* encoder_layers_11_ln1_bias {nullptr};
   void* encoder_layers_11_mha_qkv_weight {nullptr};
   void* encoder_layers_11_mha_qkv_bias {nullptr};
   void* encoder_layers_11_mha_proj_weight {nullptr};
   void* encoder_layers_11_mha_proj_bias {nullptr};
   void* encoder_layers_11_ln2_weight {nullptr};
   void* encoder_layers_11_ln2_bias {nullptr};
   void* encoder_layers_11_mlp_fc1_weight {nullptr};
   void* encoder_layers_11_mlp_fc1_bias {nullptr};
   void* encoder_layers_11_mlp_fc2_weight {nullptr};
   void* encoder_layers_11_mlp_fc2_bias {nullptr};
   void* encoder_layers_12_ln1_weight {nullptr};
   void* encoder_layers_12_ln1_bias {nullptr};
   void* encoder_layers_12_mha_qkv_weight {nullptr};
   void* encoder_layers_12_mha_qkv_bias {nullptr};
   void* encoder_layers_12_mha_proj_weight {nullptr};
   void* encoder_layers_12_mha_proj_bias {nullptr};
   void* encoder_layers_12_ln2_weight {nullptr};
   void* encoder_layers_12_ln2_bias {nullptr};
   void* encoder_layers_12_mlp_fc1_weight {nullptr};
   void* encoder_layers_12_mlp_fc1_bias {nullptr};
   void* encoder_layers_12_mlp_fc2_weight {nullptr};
   void* encoder_layers_12_mlp_fc2_bias {nullptr};
   void* encoder_layers_13_ln1_weight {nullptr};
   void* encoder_layers_13_ln1_bias {nullptr};
   void* encoder_layers_13_mha_qkv_weight {nullptr};
   void* encoder_layers_13_mha_qkv_bias {nullptr};
   void* encoder_layers_13_mha_proj_weight {nullptr};
   void* encoder_layers_13_mha_proj_bias {nullptr};
   void* encoder_layers_13_ln2_weight {nullptr};
   void* encoder_layers_13_ln2_bias {nullptr};
   void* encoder_layers_13_mlp_fc1_weight {nullptr};
   void* encoder_layers_13_mlp_fc1_bias {nullptr};
   void* encoder_layers_13_mlp_fc2_weight {nullptr};
   void* encoder_layers_13_mlp_fc2_bias {nullptr};
   void* encoder_layers_14_ln1_weight {nullptr};
   void* encoder_layers_14_ln1_bias {nullptr};
   void* encoder_layers_14_mha_qkv_weight {nullptr};
   void* encoder_layers_14_mha_qkv_bias {nullptr};
   void* encoder_layers_14_mha_proj_weight {nullptr};
   void* encoder_layers_14_mha_proj_bias {nullptr};
   void* encoder_layers_14_ln2_weight {nullptr};
   void* encoder_layers_14_ln2_bias {nullptr};
   void* encoder_layers_14_mlp_fc1_weight {nullptr};
   void* encoder_layers_14_mlp_fc1_bias {nullptr};
   void* encoder_layers_14_mlp_fc2_weight {nullptr};
   void* encoder_layers_14_mlp_fc2_bias {nullptr};
   void* encoder_layers_15_ln1_weight {nullptr};
   void* encoder_layers_15_ln1_bias {nullptr};
   void* encoder_layers_15_mha_qkv_weight {nullptr};
   void* encoder_layers_15_mha_qkv_bias {nullptr};
   void* encoder_layers_15_mha_proj_weight {nullptr};
   void* encoder_layers_15_mha_proj_bias {nullptr};
   void* encoder_layers_15_ln2_weight {nullptr};
   void* encoder_layers_15_ln2_bias {nullptr};
   void* encoder_layers_15_mlp_fc1_weight {nullptr};
   void* encoder_layers_15_mlp_fc1_bias {nullptr};
   void* encoder_layers_15_mlp_fc2_weight {nullptr};
   void* encoder_layers_15_mlp_fc2_bias {nullptr};
   void* encoder_layers_16_ln1_weight {nullptr};
   void* encoder_layers_16_ln1_bias {nullptr};
   void* encoder_layers_16_mha_qkv_weight {nullptr};
   void* encoder_layers_16_mha_qkv_bias {nullptr};
   void* encoder_layers_16_mha_proj_weight {nullptr};
   void* encoder_layers_16_mha_proj_bias {nullptr};
   void* encoder_layers_16_ln2_weight {nullptr};
   void* encoder_layers_16_ln2_bias {nullptr};
   void* encoder_layers_16_mlp_fc1_weight {nullptr};
   void* encoder_layers_16_mlp_fc1_bias {nullptr};
   void* encoder_layers_16_mlp_fc2_weight {nullptr};
   void* encoder_layers_16_mlp_fc2_bias {nullptr};
   void* encoder_layers_17_ln1_weight {nullptr};
   void* encoder_layers_17_ln1_bias {nullptr};
   void* encoder_layers_17_mha_qkv_weight {nullptr};
   void* encoder_layers_17_mha_qkv_bias {nullptr};
   void* encoder_layers_17_mha_proj_weight {nullptr};
   void* encoder_layers_17_mha_proj_bias {nullptr};
   void* encoder_layers_17_ln2_weight {nullptr};
   void* encoder_layers_17_ln2_bias {nullptr};
   void* encoder_layers_17_mlp_fc1_weight {nullptr};
   void* encoder_layers_17_mlp_fc1_bias {nullptr};
   void* encoder_layers_17_mlp_fc2_weight {nullptr};
   void* encoder_layers_17_mlp_fc2_bias {nullptr};
   void* encoder_layers_18_ln1_weight {nullptr};
   void* encoder_layers_18_ln1_bias {nullptr};
   void* encoder_layers_18_mha_qkv_weight {nullptr};
   void* encoder_layers_18_mha_qkv_bias {nullptr};
   void* encoder_layers_18_mha_proj_weight {nullptr};
   void* encoder_layers_18_mha_proj_bias {nullptr};
   void* encoder_layers_18_ln2_weight {nullptr};
   void* encoder_layers_18_ln2_bias {nullptr};
   void* encoder_layers_18_mlp_fc1_weight {nullptr};
   void* encoder_layers_18_mlp_fc1_bias {nullptr};
   void* encoder_layers_18_mlp_fc2_weight {nullptr};
   void* encoder_layers_18_mlp_fc2_bias {nullptr};
   void* encoder_layers_19_ln1_weight {nullptr};
   void* encoder_layers_19_ln1_bias {nullptr};
   void* encoder_layers_19_mha_qkv_weight {nullptr};
   void* encoder_layers_19_mha_qkv_bias {nullptr};
   void* encoder_layers_19_mha_proj_weight {nullptr};
   void* encoder_layers_19_mha_proj_bias {nullptr};
   void* encoder_layers_19_ln2_weight {nullptr};
   void* encoder_layers_19_ln2_bias {nullptr};
   void* encoder_layers_19_mlp_fc1_weight {nullptr};
   void* encoder_layers_19_mlp_fc1_bias {nullptr};
   void* encoder_layers_19_mlp_fc2_weight {nullptr};
   void* encoder_layers_19_mlp_fc2_bias {nullptr};
   void* encoder_layers_20_ln1_weight {nullptr};
   void* encoder_layers_20_ln1_bias {nullptr};
   void* encoder_layers_20_mha_qkv_weight {nullptr};
   void* encoder_layers_20_mha_qkv_bias {nullptr};
   void* encoder_layers_20_mha_proj_weight {nullptr};
   void* encoder_layers_20_mha_proj_bias {nullptr};
   void* encoder_layers_20_ln2_weight {nullptr};
   void* encoder_layers_20_ln2_bias {nullptr};
   void* encoder_layers_20_mlp_fc1_weight {nullptr};
   void* encoder_layers_20_mlp_fc1_bias {nullptr};
   void* encoder_layers_20_mlp_fc2_weight {nullptr};
   void* encoder_layers_20_mlp_fc2_bias {nullptr};
   void* encoder_layers_21_ln1_weight {nullptr};
   void* encoder_layers_21_ln1_bias {nullptr};
   void* encoder_layers_21_mha_qkv_weight {nullptr};
   void* encoder_layers_21_mha_qkv_bias {nullptr};
   void* encoder_layers_21_mha_proj_weight {nullptr};
   void* encoder_layers_21_mha_proj_bias {nullptr};
   void* encoder_layers_21_ln2_weight {nullptr};
   void* encoder_layers_21_ln2_bias {nullptr};
   void* encoder_layers_21_mlp_fc1_weight {nullptr};
   void* encoder_layers_21_mlp_fc1_bias {nullptr};
   void* encoder_layers_21_mlp_fc2_weight {nullptr};
   void* encoder_layers_21_mlp_fc2_bias {nullptr};
   void* encoder_layers_22_ln1_weight {nullptr};
   void* encoder_layers_22_ln1_bias {nullptr};
   void* encoder_layers_22_mha_qkv_weight {nullptr};
   void* encoder_layers_22_mha_qkv_bias {nullptr};
   void* encoder_layers_22_mha_proj_weight {nullptr};
   void* encoder_layers_22_mha_proj_bias {nullptr};
   void* encoder_layers_22_ln2_weight {nullptr};
   void* encoder_layers_22_ln2_bias {nullptr};
   void* encoder_layers_22_mlp_fc1_weight {nullptr};
   void* encoder_layers_22_mlp_fc1_bias {nullptr};
   void* encoder_layers_22_mlp_fc2_weight {nullptr};
   void* encoder_layers_22_mlp_fc2_bias {nullptr};
   void* encoder_layers_23_ln1_weight {nullptr};
   void* encoder_layers_23_ln1_bias {nullptr};
   void* encoder_layers_23_mha_qkv_weight {nullptr};
   void* encoder_layers_23_mha_qkv_bias {nullptr};
   void* encoder_layers_23_mha_proj_weight {nullptr};
   void* encoder_layers_23_mha_proj_bias {nullptr};
   void* encoder_layers_23_ln2_weight {nullptr};
   void* encoder_layers_23_ln2_bias {nullptr};
   void* encoder_layers_23_mlp_fc1_weight {nullptr};
   void* encoder_layers_23_mlp_fc1_bias {nullptr};
   void* encoder_layers_23_mlp_fc2_weight {nullptr};
   void* encoder_layers_23_mlp_fc2_bias {nullptr};
   void* encoder_layers_24_ln1_weight {nullptr};
   void* encoder_layers_24_ln1_bias {nullptr};
   void* encoder_layers_24_mha_qkv_weight {nullptr};
   void* encoder_layers_24_mha_qkv_bias {nullptr};
   void* encoder_layers_24_mha_proj_weight {nullptr};
   void* encoder_layers_24_mha_proj_bias {nullptr};
   void* encoder_layers_24_ln2_weight {nullptr};
   void* encoder_layers_24_ln2_bias {nullptr};
   void* encoder_layers_24_mlp_fc1_weight {nullptr};
   void* encoder_layers_24_mlp_fc1_bias {nullptr};
   void* encoder_layers_24_mlp_fc2_weight {nullptr};
   void* encoder_layers_24_mlp_fc2_bias {nullptr};
   void* encoder_layers_25_ln1_weight {nullptr};
   void* encoder_layers_25_ln1_bias {nullptr};
   void* encoder_layers_25_mha_qkv_weight {nullptr};
   void* encoder_layers_25_mha_qkv_bias {nullptr};
   void* encoder_layers_25_mha_proj_weight {nullptr};
   void* encoder_layers_25_mha_proj_bias {nullptr};
   void* encoder_layers_25_ln2_weight {nullptr};
   void* encoder_layers_25_ln2_bias {nullptr};
   void* encoder_layers_25_mlp_fc1_weight {nullptr};
   void* encoder_layers_25_mlp_fc1_bias {nullptr};
   void* encoder_layers_25_mlp_fc2_weight {nullptr};
   void* encoder_layers_25_mlp_fc2_bias {nullptr};
   void* encoder_layers_26_ln1_weight {nullptr};
   void* encoder_layers_26_ln1_bias {nullptr};
   void* encoder_layers_26_mha_qkv_weight {nullptr};
   void* encoder_layers_26_mha_qkv_bias {nullptr};
   void* encoder_layers_26_mha_proj_weight {nullptr};
   void* encoder_layers_26_mha_proj_bias {nullptr};
   void* encoder_layers_26_ln2_weight {nullptr};
   void* encoder_layers_26_ln2_bias {nullptr};
   void* encoder_layers_26_mlp_fc1_weight {nullptr};
   void* encoder_layers_26_mlp_fc1_bias {nullptr};
   void* encoder_layers_26_mlp_fc2_weight {nullptr};
   void* encoder_layers_26_mlp_fc2_bias {nullptr};
   void* encoder_ln_weight {nullptr};
   void* encoder_ln_bias {nullptr};
   void* pool_kv_weight {nullptr};
   void* pool_kv_bias {nullptr};
   void* pool_proj_weight {nullptr};
   void* pool_proj_bias {nullptr};
   void* pool_ln_weight {nullptr};
   void* pool_ln_bias {nullptr};
   void* pool_mlp_fc1_weight {nullptr};
   void* pool_mlp_fc1_bias {nullptr};
   void* pool_mlp_fc2_weight {nullptr};
   void* pool_mlp_fc2_bias {nullptr};
   void* conv2d_bias_few_channels_0_0 {nullptr};
   void* elementwise_2_0 {nullptr};
   void* reshape_4_0 {nullptr};
   void* reshape_6_0 {nullptr};
   void* split_7_0 {nullptr};
   void* split_7_1 {nullptr};
   void* split_7_2 {nullptr};
   void* reshape_8_0 {nullptr};
   void* reshape_9_0 {nullptr};
   void* reshape_10_0 {nullptr};
   void* mem_eff_attention_11_0 {nullptr};
   void* reshape_14_0 {nullptr};
   void* layernorm_15_0 {nullptr};
   void* gemm_rcr_bias_gelu_16_0 {nullptr};
   void* gemm_rcr_bias_add_17_0 {nullptr};
   void* reshape_19_0 {nullptr};
   void* reshape_21_0 {nullptr};
   void* split_22_0 {nullptr};
   void* split_22_1 {nullptr};
   void* split_22_2 {nullptr};
   void* reshape_23_0 {nullptr};
   void* reshape_24_0 {nullptr};
   void* reshape_25_0 {nullptr};
   void* mem_eff_attention_26_0 {nullptr};
   void* reshape_29_0 {nullptr};
   void* layernorm_30_0 {nullptr};
   void* gemm_rcr_bias_gelu_31_0 {nullptr};
   void* gemm_rcr_bias_add_32_0 {nullptr};
   void* reshape_34_0 {nullptr};
   void* reshape_36_0 {nullptr};
   void* split_37_0 {nullptr};
   void* split_37_1 {nullptr};
   void* split_37_2 {nullptr};
   void* reshape_38_0 {nullptr};
   void* reshape_39_0 {nullptr};
   void* reshape_40_0 {nullptr};
   void* mem_eff_attention_41_0 {nullptr};
   void* reshape_44_0 {nullptr};
   void* layernorm_45_0 {nullptr};
   void* gemm_rcr_bias_gelu_46_0 {nullptr};
   void* gemm_rcr_bias_add_47_0 {nullptr};
   void* reshape_49_0 {nullptr};
   void* reshape_51_0 {nullptr};
   void* split_52_0 {nullptr};
   void* split_52_1 {nullptr};
   void* split_52_2 {nullptr};
   void* reshape_53_0 {nullptr};
   void* reshape_54_0 {nullptr};
   void* reshape_55_0 {nullptr};
   void* mem_eff_attention_56_0 {nullptr};
   void* reshape_59_0 {nullptr};
   void* layernorm_60_0 {nullptr};
   void* gemm_rcr_bias_gelu_61_0 {nullptr};
   void* gemm_rcr_bias_add_62_0 {nullptr};
   void* reshape_64_0 {nullptr};
   void* reshape_66_0 {nullptr};
   void* split_67_0 {nullptr};
   void* split_67_1 {nullptr};
   void* split_67_2 {nullptr};
   void* reshape_68_0 {nullptr};
   void* reshape_69_0 {nullptr};
   void* reshape_70_0 {nullptr};
   void* mem_eff_attention_71_0 {nullptr};
   void* reshape_74_0 {nullptr};
   void* layernorm_75_0 {nullptr};
   void* gemm_rcr_bias_gelu_76_0 {nullptr};
   void* gemm_rcr_bias_add_77_0 {nullptr};
   void* reshape_79_0 {nullptr};
   void* reshape_81_0 {nullptr};
   void* split_82_0 {nullptr};
   void* split_82_1 {nullptr};
   void* split_82_2 {nullptr};
   void* reshape_83_0 {nullptr};
   void* reshape_84_0 {nullptr};
   void* reshape_85_0 {nullptr};
   void* mem_eff_attention_86_0 {nullptr};
   void* reshape_89_0 {nullptr};
   void* layernorm_90_0 {nullptr};
   void* gemm_rcr_bias_gelu_91_0 {nullptr};
   void* gemm_rcr_bias_add_92_0 {nullptr};
   void* reshape_94_0 {nullptr};
   void* reshape_96_0 {nullptr};
   void* split_97_0 {nullptr};
   void* split_97_1 {nullptr};
   void* split_97_2 {nullptr};
   void* reshape_98_0 {nullptr};
   void* reshape_99_0 {nullptr};
   void* reshape_100_0 {nullptr};
   void* mem_eff_attention_101_0 {nullptr};
   void* reshape_104_0 {nullptr};
   void* layernorm_105_0 {nullptr};
   void* gemm_rcr_bias_gelu_106_0 {nullptr};
   void* gemm_rcr_bias_add_107_0 {nullptr};
   void* reshape_109_0 {nullptr};
   void* reshape_111_0 {nullptr};
   void* split_112_0 {nullptr};
   void* split_112_1 {nullptr};
   void* split_112_2 {nullptr};
   void* reshape_113_0 {nullptr};
   void* reshape_114_0 {nullptr};
   void* reshape_115_0 {nullptr};
   void* mem_eff_attention_116_0 {nullptr};
   void* reshape_119_0 {nullptr};
   void* layernorm_120_0 {nullptr};
   void* gemm_rcr_bias_gelu_121_0 {nullptr};
   void* gemm_rcr_bias_add_122_0 {nullptr};
   void* reshape_124_0 {nullptr};
   void* reshape_126_0 {nullptr};
   void* split_127_0 {nullptr};
   void* split_127_1 {nullptr};
   void* split_127_2 {nullptr};
   void* reshape_128_0 {nullptr};
   void* reshape_129_0 {nullptr};
   void* reshape_130_0 {nullptr};
   void* mem_eff_attention_131_0 {nullptr};
   void* reshape_134_0 {nullptr};
   void* layernorm_135_0 {nullptr};
   void* gemm_rcr_bias_gelu_136_0 {nullptr};
   void* gemm_rcr_bias_add_137_0 {nullptr};
   void* reshape_139_0 {nullptr};
   void* reshape_141_0 {nullptr};
   void* split_142_0 {nullptr};
   void* split_142_1 {nullptr};
   void* split_142_2 {nullptr};
   void* reshape_143_0 {nullptr};
   void* reshape_144_0 {nullptr};
   void* reshape_145_0 {nullptr};
   void* mem_eff_attention_146_0 {nullptr};
   void* reshape_149_0 {nullptr};
   void* layernorm_150_0 {nullptr};
   void* gemm_rcr_bias_gelu_151_0 {nullptr};
   void* gemm_rcr_bias_add_152_0 {nullptr};
   void* reshape_154_0 {nullptr};
   void* reshape_156_0 {nullptr};
   void* split_157_0 {nullptr};
   void* split_157_1 {nullptr};
   void* split_157_2 {nullptr};
   void* reshape_158_0 {nullptr};
   void* reshape_159_0 {nullptr};
   void* reshape_160_0 {nullptr};
   void* mem_eff_attention_161_0 {nullptr};
   void* reshape_164_0 {nullptr};
   void* layernorm_165_0 {nullptr};
   void* gemm_rcr_bias_gelu_166_0 {nullptr};
   void* gemm_rcr_bias_add_167_0 {nullptr};
   void* reshape_169_0 {nullptr};
   void* reshape_171_0 {nullptr};
   void* split_172_0 {nullptr};
   void* split_172_1 {nullptr};
   void* split_172_2 {nullptr};
   void* reshape_173_0 {nullptr};
   void* reshape_174_0 {nullptr};
   void* reshape_175_0 {nullptr};
   void* mem_eff_attention_176_0 {nullptr};
   void* reshape_179_0 {nullptr};
   void* layernorm_180_0 {nullptr};
   void* gemm_rcr_bias_gelu_181_0 {nullptr};
   void* gemm_rcr_bias_add_182_0 {nullptr};
   void* reshape_184_0 {nullptr};
   void* reshape_186_0 {nullptr};
   void* split_187_0 {nullptr};
   void* split_187_1 {nullptr};
   void* split_187_2 {nullptr};
   void* reshape_188_0 {nullptr};
   void* reshape_189_0 {nullptr};
   void* reshape_190_0 {nullptr};
   void* mem_eff_attention_191_0 {nullptr};
   void* reshape_194_0 {nullptr};
   void* layernorm_195_0 {nullptr};
   void* gemm_rcr_bias_gelu_196_0 {nullptr};
   void* gemm_rcr_bias_add_197_0 {nullptr};
   void* reshape_199_0 {nullptr};
   void* reshape_201_0 {nullptr};
   void* split_202_0 {nullptr};
   void* split_202_1 {nullptr};
   void* split_202_2 {nullptr};
   void* reshape_203_0 {nullptr};
   void* reshape_204_0 {nullptr};
   void* reshape_205_0 {nullptr};
   void* mem_eff_attention_206_0 {nullptr};
   void* reshape_209_0 {nullptr};
   void* layernorm_210_0 {nullptr};
   void* gemm_rcr_bias_gelu_211_0 {nullptr};
   void* gemm_rcr_bias_add_212_0 {nullptr};
   void* reshape_214_0 {nullptr};
   void* reshape_216_0 {nullptr};
   void* split_217_0 {nullptr};
   void* split_217_1 {nullptr};
   void* split_217_2 {nullptr};
   void* reshape_218_0 {nullptr};
   void* reshape_219_0 {nullptr};
   void* reshape_220_0 {nullptr};
   void* mem_eff_attention_221_0 {nullptr};
   void* reshape_224_0 {nullptr};
   void* layernorm_225_0 {nullptr};
   void* gemm_rcr_bias_gelu_226_0 {nullptr};
   void* gemm_rcr_bias_add_227_0 {nullptr};
   void* reshape_229_0 {nullptr};
   void* reshape_231_0 {nullptr};
   void* split_232_0 {nullptr};
   void* split_232_1 {nullptr};
   void* split_232_2 {nullptr};
   void* reshape_233_0 {nullptr};
   void* reshape_234_0 {nullptr};
   void* reshape_235_0 {nullptr};
   void* mem_eff_attention_236_0 {nullptr};
   void* reshape_239_0 {nullptr};
   void* layernorm_240_0 {nullptr};
   void* gemm_rcr_bias_gelu_241_0 {nullptr};
   void* gemm_rcr_bias_add_242_0 {nullptr};
   void* reshape_244_0 {nullptr};
   void* reshape_246_0 {nullptr};
   void* split_247_0 {nullptr};
   void* split_247_1 {nullptr};
   void* split_247_2 {nullptr};
   void* reshape_248_0 {nullptr};
   void* reshape_249_0 {nullptr};
   void* reshape_250_0 {nullptr};
   void* mem_eff_attention_251_0 {nullptr};
   void* reshape_254_0 {nullptr};
   void* layernorm_255_0 {nullptr};
   void* gemm_rcr_bias_gelu_256_0 {nullptr};
   void* gemm_rcr_bias_add_257_0 {nullptr};
   void* reshape_259_0 {nullptr};
   void* reshape_261_0 {nullptr};
   void* split_262_0 {nullptr};
   void* split_262_1 {nullptr};
   void* split_262_2 {nullptr};
   void* reshape_263_0 {nullptr};
   void* reshape_264_0 {nullptr};
   void* reshape_265_0 {nullptr};
   void* mem_eff_attention_266_0 {nullptr};
   void* reshape_269_0 {nullptr};
   void* layernorm_270_0 {nullptr};
   void* gemm_rcr_bias_gelu_271_0 {nullptr};
   void* gemm_rcr_bias_add_272_0 {nullptr};
   void* reshape_274_0 {nullptr};
   void* reshape_276_0 {nullptr};
   void* split_277_0 {nullptr};
   void* split_277_1 {nullptr};
   void* split_277_2 {nullptr};
   void* reshape_278_0 {nullptr};
   void* reshape_279_0 {nullptr};
   void* reshape_280_0 {nullptr};
   void* mem_eff_attention_281_0 {nullptr};
   void* reshape_284_0 {nullptr};
   void* layernorm_285_0 {nullptr};
   void* gemm_rcr_bias_gelu_286_0 {nullptr};
   void* gemm_rcr_bias_add_287_0 {nullptr};
   void* reshape_289_0 {nullptr};
   void* reshape_291_0 {nullptr};
   void* split_292_0 {nullptr};
   void* split_292_1 {nullptr};
   void* split_292_2 {nullptr};
   void* reshape_293_0 {nullptr};
   void* reshape_294_0 {nullptr};
   void* reshape_295_0 {nullptr};
   void* mem_eff_attention_296_0 {nullptr};
   void* reshape_299_0 {nullptr};
   void* layernorm_300_0 {nullptr};
   void* gemm_rcr_bias_gelu_301_0 {nullptr};
   void* gemm_rcr_bias_add_302_0 {nullptr};
   void* reshape_304_0 {nullptr};
   void* reshape_306_0 {nullptr};
   void* split_307_0 {nullptr};
   void* split_307_1 {nullptr};
   void* split_307_2 {nullptr};
   void* reshape_308_0 {nullptr};
   void* reshape_309_0 {nullptr};
   void* reshape_310_0 {nullptr};
   void* mem_eff_attention_311_0 {nullptr};
   void* reshape_314_0 {nullptr};
   void* layernorm_315_0 {nullptr};
   void* gemm_rcr_bias_gelu_316_0 {nullptr};
   void* gemm_rcr_bias_add_317_0 {nullptr};
   void* reshape_319_0 {nullptr};
   void* reshape_321_0 {nullptr};
   void* split_322_0 {nullptr};
   void* split_322_1 {nullptr};
   void* split_322_2 {nullptr};
   void* reshape_323_0 {nullptr};
   void* reshape_324_0 {nullptr};
   void* reshape_325_0 {nullptr};
   void* mem_eff_attention_326_0 {nullptr};
   void* reshape_329_0 {nullptr};
   void* layernorm_330_0 {nullptr};
   void* gemm_rcr_bias_gelu_331_0 {nullptr};
   void* gemm_rcr_bias_add_332_0 {nullptr};
   void* reshape_334_0 {nullptr};
   void* reshape_336_0 {nullptr};
   void* split_337_0 {nullptr};
   void* split_337_1 {nullptr};
   void* split_337_2 {nullptr};
   void* reshape_338_0 {nullptr};
   void* reshape_339_0 {nullptr};
   void* reshape_340_0 {nullptr};
   void* mem_eff_attention_341_0 {nullptr};
   void* reshape_344_0 {nullptr};
   void* layernorm_345_0 {nullptr};
   void* gemm_rcr_bias_gelu_346_0 {nullptr};
   void* gemm_rcr_bias_add_347_0 {nullptr};
   void* reshape_349_0 {nullptr};
   void* reshape_351_0 {nullptr};
   void* split_352_0 {nullptr};
   void* split_352_1 {nullptr};
   void* split_352_2 {nullptr};
   void* reshape_353_0 {nullptr};
   void* reshape_354_0 {nullptr};
   void* reshape_355_0 {nullptr};
   void* mem_eff_attention_356_0 {nullptr};
   void* reshape_359_0 {nullptr};
   void* layernorm_360_0 {nullptr};
   void* gemm_rcr_bias_gelu_361_0 {nullptr};
   void* gemm_rcr_bias_add_362_0 {nullptr};
   void* reshape_364_0 {nullptr};
   void* reshape_366_0 {nullptr};
   void* split_367_0 {nullptr};
   void* split_367_1 {nullptr};
   void* split_367_2 {nullptr};
   void* reshape_368_0 {nullptr};
   void* reshape_369_0 {nullptr};
   void* reshape_370_0 {nullptr};
   void* mem_eff_attention_371_0 {nullptr};
   void* reshape_374_0 {nullptr};
   void* layernorm_375_0 {nullptr};
   void* gemm_rcr_bias_gelu_376_0 {nullptr};
   void* gemm_rcr_bias_add_377_0 {nullptr};
   void* reshape_379_0 {nullptr};
   void* reshape_381_0 {nullptr};
   void* split_382_0 {nullptr};
   void* split_382_1 {nullptr};
   void* split_382_2 {nullptr};
   void* reshape_383_0 {nullptr};
   void* reshape_384_0 {nullptr};
   void* reshape_385_0 {nullptr};
   void* mem_eff_attention_386_0 {nullptr};
   void* reshape_389_0 {nullptr};
   void* layernorm_390_0 {nullptr};
   void* gemm_rcr_bias_gelu_391_0 {nullptr};
   void* gemm_rcr_bias_add_392_0 {nullptr};
   void* reshape_394_0 {nullptr};
   void* reshape_396_0 {nullptr};
   void* split_397_0 {nullptr};
   void* split_397_1 {nullptr};
   void* split_397_2 {nullptr};
   void* reshape_398_0 {nullptr};
   void* reshape_399_0 {nullptr};
   void* reshape_400_0 {nullptr};
   void* mem_eff_attention_401_0 {nullptr};
   void* reshape_404_0 {nullptr};
   void* layernorm_405_0 {nullptr};
   void* gemm_rcr_bias_gelu_406_0 {nullptr};
   void* gemm_rcr_bias_add_407_0 {nullptr};
   void* layernorm_408_0 {nullptr};
   void* reshape_410_0 {nullptr};
   void* permute_411_0 {nullptr};
   void* split_412_0 {nullptr};
   void* split_412_1 {nullptr};
   void* squeeze_413_0 {nullptr};
   void* squeeze_414_0 {nullptr};
   void* reshape_417_0 {nullptr};
   void* mem_eff_attention_418_0 {nullptr};
   void* reshape_426_0 {nullptr};
   void* gemm_rcr_bias_421_0 {nullptr};
   void* layernorm_422_0 {nullptr};
   void* gemm_rcr_bias_gelu_423_0 {nullptr};
   void* output_0 {nullptr};
   int64_t pool_probe_dim_0 { 1 };
   int64_t pool_probe_dim_1 { 1 };
   int64_t pool_probe_dim_2 { 1152 };
   int64_t pool_q_weight_dim_0 { 1152 };
   int64_t pool_q_weight_dim_1 { 1152 };
   int64_t pool_q_bias_dim_0 { 1152 };
   int64_t input0_dim_0 { 8 };
   int64_t input0_dim_1 { 384 };
   int64_t input0_dim_2 { 384 };
   int64_t input0_dim_3 { 3 };
   int64_t patch_embed_proj_weight_dim_0 { 1152 };
   int64_t patch_embed_proj_weight_dim_1 { 14 };
   int64_t patch_embed_proj_weight_dim_2 { 14 };
   int64_t patch_embed_proj_weight_dim_3 { 3 };
   int64_t patch_embed_proj_bias_dim_0 { 1152 };
   int64_t pos_emb_pos_emb_dim_0 { 1 };
   int64_t pos_emb_pos_emb_dim_1 { 729 };
   int64_t pos_emb_pos_emb_dim_2 { 1152 };
   int64_t encoder_layers_0_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_0_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_0_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_0_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_0_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_0_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_0_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_0_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_0_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_0_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_0_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_0_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_0_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_0_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_0_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_0_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_1_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_1_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_1_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_1_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_1_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_1_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_1_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_1_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_1_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_1_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_1_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_1_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_1_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_1_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_1_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_1_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_2_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_2_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_2_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_2_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_2_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_2_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_2_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_2_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_2_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_2_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_2_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_2_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_2_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_2_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_2_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_2_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_3_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_3_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_3_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_3_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_3_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_3_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_3_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_3_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_3_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_3_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_3_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_3_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_3_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_3_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_3_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_3_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_4_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_4_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_4_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_4_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_4_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_4_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_4_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_4_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_4_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_4_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_4_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_4_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_4_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_4_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_4_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_4_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_5_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_5_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_5_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_5_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_5_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_5_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_5_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_5_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_5_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_5_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_5_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_5_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_5_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_5_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_5_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_5_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_6_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_6_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_6_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_6_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_6_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_6_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_6_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_6_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_6_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_6_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_6_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_6_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_6_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_6_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_6_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_6_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_7_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_7_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_7_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_7_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_7_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_7_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_7_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_7_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_7_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_7_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_7_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_7_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_7_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_7_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_7_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_7_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_8_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_8_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_8_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_8_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_8_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_8_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_8_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_8_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_8_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_8_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_8_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_8_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_8_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_8_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_8_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_8_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_9_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_9_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_9_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_9_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_9_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_9_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_9_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_9_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_9_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_9_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_9_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_9_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_9_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_9_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_9_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_9_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_10_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_10_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_10_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_10_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_10_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_10_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_10_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_10_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_10_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_10_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_10_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_10_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_10_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_10_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_10_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_10_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_11_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_11_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_11_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_11_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_11_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_11_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_11_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_11_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_11_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_11_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_11_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_11_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_11_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_11_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_11_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_11_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_12_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_12_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_12_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_12_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_12_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_12_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_12_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_12_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_12_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_12_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_12_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_12_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_12_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_12_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_12_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_12_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_13_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_13_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_13_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_13_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_13_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_13_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_13_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_13_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_13_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_13_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_13_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_13_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_13_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_13_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_13_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_13_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_14_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_14_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_14_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_14_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_14_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_14_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_14_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_14_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_14_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_14_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_14_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_14_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_14_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_14_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_14_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_14_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_15_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_15_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_15_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_15_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_15_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_15_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_15_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_15_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_15_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_15_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_15_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_15_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_15_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_15_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_15_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_15_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_16_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_16_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_16_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_16_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_16_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_16_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_16_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_16_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_16_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_16_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_16_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_16_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_16_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_16_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_16_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_16_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_17_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_17_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_17_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_17_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_17_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_17_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_17_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_17_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_17_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_17_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_17_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_17_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_17_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_17_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_17_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_17_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_18_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_18_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_18_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_18_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_18_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_18_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_18_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_18_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_18_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_18_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_18_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_18_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_18_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_18_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_18_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_18_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_19_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_19_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_19_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_19_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_19_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_19_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_19_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_19_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_19_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_19_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_19_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_19_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_19_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_19_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_19_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_19_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_20_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_20_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_20_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_20_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_20_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_20_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_20_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_20_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_20_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_20_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_20_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_20_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_20_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_20_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_20_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_20_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_21_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_21_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_21_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_21_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_21_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_21_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_21_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_21_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_21_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_21_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_21_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_21_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_21_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_21_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_21_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_21_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_22_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_22_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_22_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_22_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_22_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_22_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_22_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_22_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_22_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_22_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_22_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_22_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_22_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_22_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_22_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_22_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_23_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_23_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_23_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_23_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_23_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_23_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_23_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_23_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_23_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_23_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_23_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_23_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_23_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_23_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_23_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_23_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_24_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_24_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_24_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_24_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_24_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_24_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_24_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_24_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_24_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_24_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_24_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_24_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_24_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_24_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_24_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_24_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_25_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_25_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_25_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_25_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_25_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_25_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_25_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_25_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_25_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_25_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_25_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_25_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_25_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_25_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_25_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_25_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_layers_26_ln1_weight_dim_0 { 1152 };
   int64_t encoder_layers_26_ln1_bias_dim_0 { 1152 };
   int64_t encoder_layers_26_mha_qkv_weight_dim_0 { 3456 };
   int64_t encoder_layers_26_mha_qkv_weight_dim_1 { 1152 };
   int64_t encoder_layers_26_mha_qkv_bias_dim_0 { 3456 };
   int64_t encoder_layers_26_mha_proj_weight_dim_0 { 1152 };
   int64_t encoder_layers_26_mha_proj_weight_dim_1 { 1152 };
   int64_t encoder_layers_26_mha_proj_bias_dim_0 { 1152 };
   int64_t encoder_layers_26_ln2_weight_dim_0 { 1152 };
   int64_t encoder_layers_26_ln2_bias_dim_0 { 1152 };
   int64_t encoder_layers_26_mlp_fc1_weight_dim_0 { 4304 };
   int64_t encoder_layers_26_mlp_fc1_weight_dim_1 { 1152 };
   int64_t encoder_layers_26_mlp_fc1_bias_dim_0 { 4304 };
   int64_t encoder_layers_26_mlp_fc2_weight_dim_0 { 1152 };
   int64_t encoder_layers_26_mlp_fc2_weight_dim_1 { 4304 };
   int64_t encoder_layers_26_mlp_fc2_bias_dim_0 { 1152 };
   int64_t encoder_ln_weight_dim_0 { 1152 };
   int64_t encoder_ln_bias_dim_0 { 1152 };
   int64_t pool_kv_weight_dim_0 { 2304 };
   int64_t pool_kv_weight_dim_1 { 1152 };
   int64_t pool_kv_bias_dim_0 { 2304 };
   int64_t pool_proj_weight_dim_0 { 1152 };
   int64_t pool_proj_weight_dim_1 { 1152 };
   int64_t pool_proj_bias_dim_0 { 1152 };
   int64_t pool_ln_weight_dim_0 { 1152 };
   int64_t pool_ln_bias_dim_0 { 1152 };
   int64_t pool_mlp_fc1_weight_dim_0 { 4304 };
   int64_t pool_mlp_fc1_weight_dim_1 { 1152 };
   int64_t pool_mlp_fc1_bias_dim_0 { 4304 };
   int64_t pool_mlp_fc2_weight_dim_0 { 1152 };
   int64_t pool_mlp_fc2_weight_dim_1 { 4304 };
   int64_t pool_mlp_fc2_bias_dim_0 { 1152 };
   int64_t conv2d_bias_few_channels_0_0_dim_1 { 27 };
   int64_t conv2d_bias_few_channels_0_0_dim_2 { 27 };
   int64_t conv2d_bias_few_channels_0_0_dim_3 { 1152 };
   int64_t reshape_1_0_dim_0 { 8 };
   int64_t reshape_1_0_dim_1 { 729 };
   int64_t reshape_1_0_dim_2 { 1152 };
   int64_t reshape_4_0_dim_0 { 5832 };
   int64_t reshape_4_0_dim_1 { 1152 };
   int64_t reshape_6_0_dim_0 { 3 };
   int64_t reshape_6_0_dim_1 { 8 };
   int64_t reshape_6_0_dim_2 { 16 };
   int64_t reshape_6_0_dim_3 { 729 };
   int64_t reshape_6_0_dim_4 { 72 };
   int64_t split_7_0_dim_0 { 1 };
   int64_t split_7_1_dim_0 { 1 };
   int64_t split_7_2_dim_0 { 1 };
   int64_t reshape_8_0_dim_0 { 8 };
   int64_t reshape_8_0_dim_1 { 16 };
   int64_t reshape_8_0_dim_2 { 729 };
   int64_t reshape_8_0_dim_3 { 72 };
   int64_t reshape_9_0_dim_0 { 8 };
   int64_t reshape_9_0_dim_1 { 16 };
   int64_t reshape_9_0_dim_2 { 729 };
   int64_t reshape_9_0_dim_3 { 72 };
   int64_t reshape_10_0_dim_0 { 8 };
   int64_t reshape_10_0_dim_1 { 16 };
   int64_t reshape_10_0_dim_2 { 729 };
   int64_t reshape_10_0_dim_3 { 72 };
   int64_t reshape_14_0_dim_0 { 8 };
   int64_t reshape_14_0_dim_1 { 729 };
   int64_t reshape_14_0_dim_2 { 1152 };
   int64_t reshape_12_0_dim_0 { 5832 };
   int64_t reshape_12_0_dim_1 { 1152 };
   int64_t reshape_19_0_dim_0 { 5832 };
   int64_t reshape_19_0_dim_1 { 1152 };
   int64_t reshape_21_0_dim_0 { 3 };
   int64_t reshape_21_0_dim_1 { 8 };
   int64_t reshape_21_0_dim_2 { 16 };
   int64_t reshape_21_0_dim_3 { 729 };
   int64_t reshape_21_0_dim_4 { 72 };
   int64_t split_22_0_dim_0 { 1 };
   int64_t split_22_1_dim_0 { 1 };
   int64_t split_22_2_dim_0 { 1 };
   int64_t reshape_23_0_dim_0 { 8 };
   int64_t reshape_23_0_dim_1 { 16 };
   int64_t reshape_23_0_dim_2 { 729 };
   int64_t reshape_23_0_dim_3 { 72 };
   int64_t reshape_24_0_dim_0 { 8 };
   int64_t reshape_24_0_dim_1 { 16 };
   int64_t reshape_24_0_dim_2 { 729 };
   int64_t reshape_24_0_dim_3 { 72 };
   int64_t reshape_25_0_dim_0 { 8 };
   int64_t reshape_25_0_dim_1 { 16 };
   int64_t reshape_25_0_dim_2 { 729 };
   int64_t reshape_25_0_dim_3 { 72 };
   int64_t reshape_29_0_dim_0 { 8 };
   int64_t reshape_29_0_dim_1 { 729 };
   int64_t reshape_29_0_dim_2 { 1152 };
   int64_t reshape_27_0_dim_0 { 5832 };
   int64_t reshape_27_0_dim_1 { 1152 };
   int64_t reshape_34_0_dim_0 { 5832 };
   int64_t reshape_34_0_dim_1 { 1152 };
   int64_t reshape_36_0_dim_0 { 3 };
   int64_t reshape_36_0_dim_1 { 8 };
   int64_t reshape_36_0_dim_2 { 16 };
   int64_t reshape_36_0_dim_3 { 729 };
   int64_t reshape_36_0_dim_4 { 72 };
   int64_t split_37_0_dim_0 { 1 };
   int64_t split_37_1_dim_0 { 1 };
   int64_t split_37_2_dim_0 { 1 };
   int64_t reshape_38_0_dim_0 { 8 };
   int64_t reshape_38_0_dim_1 { 16 };
   int64_t reshape_38_0_dim_2 { 729 };
   int64_t reshape_38_0_dim_3 { 72 };
   int64_t reshape_39_0_dim_0 { 8 };
   int64_t reshape_39_0_dim_1 { 16 };
   int64_t reshape_39_0_dim_2 { 729 };
   int64_t reshape_39_0_dim_3 { 72 };
   int64_t reshape_40_0_dim_0 { 8 };
   int64_t reshape_40_0_dim_1 { 16 };
   int64_t reshape_40_0_dim_2 { 729 };
   int64_t reshape_40_0_dim_3 { 72 };
   int64_t reshape_44_0_dim_0 { 8 };
   int64_t reshape_44_0_dim_1 { 729 };
   int64_t reshape_44_0_dim_2 { 1152 };
   int64_t reshape_42_0_dim_0 { 5832 };
   int64_t reshape_42_0_dim_1 { 1152 };
   int64_t reshape_49_0_dim_0 { 5832 };
   int64_t reshape_49_0_dim_1 { 1152 };
   int64_t reshape_51_0_dim_0 { 3 };
   int64_t reshape_51_0_dim_1 { 8 };
   int64_t reshape_51_0_dim_2 { 16 };
   int64_t reshape_51_0_dim_3 { 729 };
   int64_t reshape_51_0_dim_4 { 72 };
   int64_t split_52_0_dim_0 { 1 };
   int64_t split_52_1_dim_0 { 1 };
   int64_t split_52_2_dim_0 { 1 };
   int64_t reshape_53_0_dim_0 { 8 };
   int64_t reshape_53_0_dim_1 { 16 };
   int64_t reshape_53_0_dim_2 { 729 };
   int64_t reshape_53_0_dim_3 { 72 };
   int64_t reshape_54_0_dim_0 { 8 };
   int64_t reshape_54_0_dim_1 { 16 };
   int64_t reshape_54_0_dim_2 { 729 };
   int64_t reshape_54_0_dim_3 { 72 };
   int64_t reshape_55_0_dim_0 { 8 };
   int64_t reshape_55_0_dim_1 { 16 };
   int64_t reshape_55_0_dim_2 { 729 };
   int64_t reshape_55_0_dim_3 { 72 };
   int64_t reshape_59_0_dim_0 { 8 };
   int64_t reshape_59_0_dim_1 { 729 };
   int64_t reshape_59_0_dim_2 { 1152 };
   int64_t reshape_57_0_dim_0 { 5832 };
   int64_t reshape_57_0_dim_1 { 1152 };
   int64_t reshape_64_0_dim_0 { 5832 };
   int64_t reshape_64_0_dim_1 { 1152 };
   int64_t reshape_66_0_dim_0 { 3 };
   int64_t reshape_66_0_dim_1 { 8 };
   int64_t reshape_66_0_dim_2 { 16 };
   int64_t reshape_66_0_dim_3 { 729 };
   int64_t reshape_66_0_dim_4 { 72 };
   int64_t split_67_0_dim_0 { 1 };
   int64_t split_67_1_dim_0 { 1 };
   int64_t split_67_2_dim_0 { 1 };
   int64_t reshape_68_0_dim_0 { 8 };
   int64_t reshape_68_0_dim_1 { 16 };
   int64_t reshape_68_0_dim_2 { 729 };
   int64_t reshape_68_0_dim_3 { 72 };
   int64_t reshape_69_0_dim_0 { 8 };
   int64_t reshape_69_0_dim_1 { 16 };
   int64_t reshape_69_0_dim_2 { 729 };
   int64_t reshape_69_0_dim_3 { 72 };
   int64_t reshape_70_0_dim_0 { 8 };
   int64_t reshape_70_0_dim_1 { 16 };
   int64_t reshape_70_0_dim_2 { 729 };
   int64_t reshape_70_0_dim_3 { 72 };
   int64_t reshape_74_0_dim_0 { 8 };
   int64_t reshape_74_0_dim_1 { 729 };
   int64_t reshape_74_0_dim_2 { 1152 };
   int64_t reshape_72_0_dim_0 { 5832 };
   int64_t reshape_72_0_dim_1 { 1152 };
   int64_t reshape_79_0_dim_0 { 5832 };
   int64_t reshape_79_0_dim_1 { 1152 };
   int64_t reshape_81_0_dim_0 { 3 };
   int64_t reshape_81_0_dim_1 { 8 };
   int64_t reshape_81_0_dim_2 { 16 };
   int64_t reshape_81_0_dim_3 { 729 };
   int64_t reshape_81_0_dim_4 { 72 };
   int64_t split_82_0_dim_0 { 1 };
   int64_t split_82_1_dim_0 { 1 };
   int64_t split_82_2_dim_0 { 1 };
   int64_t reshape_83_0_dim_0 { 8 };
   int64_t reshape_83_0_dim_1 { 16 };
   int64_t reshape_83_0_dim_2 { 729 };
   int64_t reshape_83_0_dim_3 { 72 };
   int64_t reshape_84_0_dim_0 { 8 };
   int64_t reshape_84_0_dim_1 { 16 };
   int64_t reshape_84_0_dim_2 { 729 };
   int64_t reshape_84_0_dim_3 { 72 };
   int64_t reshape_85_0_dim_0 { 8 };
   int64_t reshape_85_0_dim_1 { 16 };
   int64_t reshape_85_0_dim_2 { 729 };
   int64_t reshape_85_0_dim_3 { 72 };
   int64_t reshape_89_0_dim_0 { 8 };
   int64_t reshape_89_0_dim_1 { 729 };
   int64_t reshape_89_0_dim_2 { 1152 };
   int64_t reshape_87_0_dim_0 { 5832 };
   int64_t reshape_87_0_dim_1 { 1152 };
   int64_t reshape_94_0_dim_0 { 5832 };
   int64_t reshape_94_0_dim_1 { 1152 };
   int64_t reshape_96_0_dim_0 { 3 };
   int64_t reshape_96_0_dim_1 { 8 };
   int64_t reshape_96_0_dim_2 { 16 };
   int64_t reshape_96_0_dim_3 { 729 };
   int64_t reshape_96_0_dim_4 { 72 };
   int64_t split_97_0_dim_0 { 1 };
   int64_t split_97_1_dim_0 { 1 };
   int64_t split_97_2_dim_0 { 1 };
   int64_t reshape_98_0_dim_0 { 8 };
   int64_t reshape_98_0_dim_1 { 16 };
   int64_t reshape_98_0_dim_2 { 729 };
   int64_t reshape_98_0_dim_3 { 72 };
   int64_t reshape_99_0_dim_0 { 8 };
   int64_t reshape_99_0_dim_1 { 16 };
   int64_t reshape_99_0_dim_2 { 729 };
   int64_t reshape_99_0_dim_3 { 72 };
   int64_t reshape_100_0_dim_0 { 8 };
   int64_t reshape_100_0_dim_1 { 16 };
   int64_t reshape_100_0_dim_2 { 729 };
   int64_t reshape_100_0_dim_3 { 72 };
   int64_t reshape_104_0_dim_0 { 8 };
   int64_t reshape_104_0_dim_1 { 729 };
   int64_t reshape_104_0_dim_2 { 1152 };
   int64_t reshape_102_0_dim_0 { 5832 };
   int64_t reshape_102_0_dim_1 { 1152 };
   int64_t reshape_109_0_dim_0 { 5832 };
   int64_t reshape_109_0_dim_1 { 1152 };
   int64_t reshape_111_0_dim_0 { 3 };
   int64_t reshape_111_0_dim_1 { 8 };
   int64_t reshape_111_0_dim_2 { 16 };
   int64_t reshape_111_0_dim_3 { 729 };
   int64_t reshape_111_0_dim_4 { 72 };
   int64_t split_112_0_dim_0 { 1 };
   int64_t split_112_1_dim_0 { 1 };
   int64_t split_112_2_dim_0 { 1 };
   int64_t reshape_113_0_dim_0 { 8 };
   int64_t reshape_113_0_dim_1 { 16 };
   int64_t reshape_113_0_dim_2 { 729 };
   int64_t reshape_113_0_dim_3 { 72 };
   int64_t reshape_114_0_dim_0 { 8 };
   int64_t reshape_114_0_dim_1 { 16 };
   int64_t reshape_114_0_dim_2 { 729 };
   int64_t reshape_114_0_dim_3 { 72 };
   int64_t reshape_115_0_dim_0 { 8 };
   int64_t reshape_115_0_dim_1 { 16 };
   int64_t reshape_115_0_dim_2 { 729 };
   int64_t reshape_115_0_dim_3 { 72 };
   int64_t reshape_119_0_dim_0 { 8 };
   int64_t reshape_119_0_dim_1 { 729 };
   int64_t reshape_119_0_dim_2 { 1152 };
   int64_t reshape_117_0_dim_0 { 5832 };
   int64_t reshape_117_0_dim_1 { 1152 };
   int64_t reshape_124_0_dim_0 { 5832 };
   int64_t reshape_124_0_dim_1 { 1152 };
   int64_t reshape_126_0_dim_0 { 3 };
   int64_t reshape_126_0_dim_1 { 8 };
   int64_t reshape_126_0_dim_2 { 16 };
   int64_t reshape_126_0_dim_3 { 729 };
   int64_t reshape_126_0_dim_4 { 72 };
   int64_t split_127_0_dim_0 { 1 };
   int64_t split_127_1_dim_0 { 1 };
   int64_t split_127_2_dim_0 { 1 };
   int64_t reshape_128_0_dim_0 { 8 };
   int64_t reshape_128_0_dim_1 { 16 };
   int64_t reshape_128_0_dim_2 { 729 };
   int64_t reshape_128_0_dim_3 { 72 };
   int64_t reshape_129_0_dim_0 { 8 };
   int64_t reshape_129_0_dim_1 { 16 };
   int64_t reshape_129_0_dim_2 { 729 };
   int64_t reshape_129_0_dim_3 { 72 };
   int64_t reshape_130_0_dim_0 { 8 };
   int64_t reshape_130_0_dim_1 { 16 };
   int64_t reshape_130_0_dim_2 { 729 };
   int64_t reshape_130_0_dim_3 { 72 };
   int64_t reshape_134_0_dim_0 { 8 };
   int64_t reshape_134_0_dim_1 { 729 };
   int64_t reshape_134_0_dim_2 { 1152 };
   int64_t reshape_132_0_dim_0 { 5832 };
   int64_t reshape_132_0_dim_1 { 1152 };
   int64_t reshape_139_0_dim_0 { 5832 };
   int64_t reshape_139_0_dim_1 { 1152 };
   int64_t reshape_141_0_dim_0 { 3 };
   int64_t reshape_141_0_dim_1 { 8 };
   int64_t reshape_141_0_dim_2 { 16 };
   int64_t reshape_141_0_dim_3 { 729 };
   int64_t reshape_141_0_dim_4 { 72 };
   int64_t split_142_0_dim_0 { 1 };
   int64_t split_142_1_dim_0 { 1 };
   int64_t split_142_2_dim_0 { 1 };
   int64_t reshape_143_0_dim_0 { 8 };
   int64_t reshape_143_0_dim_1 { 16 };
   int64_t reshape_143_0_dim_2 { 729 };
   int64_t reshape_143_0_dim_3 { 72 };
   int64_t reshape_144_0_dim_0 { 8 };
   int64_t reshape_144_0_dim_1 { 16 };
   int64_t reshape_144_0_dim_2 { 729 };
   int64_t reshape_144_0_dim_3 { 72 };
   int64_t reshape_145_0_dim_0 { 8 };
   int64_t reshape_145_0_dim_1 { 16 };
   int64_t reshape_145_0_dim_2 { 729 };
   int64_t reshape_145_0_dim_3 { 72 };
   int64_t reshape_149_0_dim_0 { 8 };
   int64_t reshape_149_0_dim_1 { 729 };
   int64_t reshape_149_0_dim_2 { 1152 };
   int64_t reshape_147_0_dim_0 { 5832 };
   int64_t reshape_147_0_dim_1 { 1152 };
   int64_t reshape_154_0_dim_0 { 5832 };
   int64_t reshape_154_0_dim_1 { 1152 };
   int64_t reshape_156_0_dim_0 { 3 };
   int64_t reshape_156_0_dim_1 { 8 };
   int64_t reshape_156_0_dim_2 { 16 };
   int64_t reshape_156_0_dim_3 { 729 };
   int64_t reshape_156_0_dim_4 { 72 };
   int64_t split_157_0_dim_0 { 1 };
   int64_t split_157_1_dim_0 { 1 };
   int64_t split_157_2_dim_0 { 1 };
   int64_t reshape_158_0_dim_0 { 8 };
   int64_t reshape_158_0_dim_1 { 16 };
   int64_t reshape_158_0_dim_2 { 729 };
   int64_t reshape_158_0_dim_3 { 72 };
   int64_t reshape_159_0_dim_0 { 8 };
   int64_t reshape_159_0_dim_1 { 16 };
   int64_t reshape_159_0_dim_2 { 729 };
   int64_t reshape_159_0_dim_3 { 72 };
   int64_t reshape_160_0_dim_0 { 8 };
   int64_t reshape_160_0_dim_1 { 16 };
   int64_t reshape_160_0_dim_2 { 729 };
   int64_t reshape_160_0_dim_3 { 72 };
   int64_t reshape_164_0_dim_0 { 8 };
   int64_t reshape_164_0_dim_1 { 729 };
   int64_t reshape_164_0_dim_2 { 1152 };
   int64_t reshape_162_0_dim_0 { 5832 };
   int64_t reshape_162_0_dim_1 { 1152 };
   int64_t reshape_169_0_dim_0 { 5832 };
   int64_t reshape_169_0_dim_1 { 1152 };
   int64_t reshape_171_0_dim_0 { 3 };
   int64_t reshape_171_0_dim_1 { 8 };
   int64_t reshape_171_0_dim_2 { 16 };
   int64_t reshape_171_0_dim_3 { 729 };
   int64_t reshape_171_0_dim_4 { 72 };
   int64_t split_172_0_dim_0 { 1 };
   int64_t split_172_1_dim_0 { 1 };
   int64_t split_172_2_dim_0 { 1 };
   int64_t reshape_173_0_dim_0 { 8 };
   int64_t reshape_173_0_dim_1 { 16 };
   int64_t reshape_173_0_dim_2 { 729 };
   int64_t reshape_173_0_dim_3 { 72 };
   int64_t reshape_174_0_dim_0 { 8 };
   int64_t reshape_174_0_dim_1 { 16 };
   int64_t reshape_174_0_dim_2 { 729 };
   int64_t reshape_174_0_dim_3 { 72 };
   int64_t reshape_175_0_dim_0 { 8 };
   int64_t reshape_175_0_dim_1 { 16 };
   int64_t reshape_175_0_dim_2 { 729 };
   int64_t reshape_175_0_dim_3 { 72 };
   int64_t reshape_179_0_dim_0 { 8 };
   int64_t reshape_179_0_dim_1 { 729 };
   int64_t reshape_179_0_dim_2 { 1152 };
   int64_t reshape_177_0_dim_0 { 5832 };
   int64_t reshape_177_0_dim_1 { 1152 };
   int64_t reshape_184_0_dim_0 { 5832 };
   int64_t reshape_184_0_dim_1 { 1152 };
   int64_t reshape_186_0_dim_0 { 3 };
   int64_t reshape_186_0_dim_1 { 8 };
   int64_t reshape_186_0_dim_2 { 16 };
   int64_t reshape_186_0_dim_3 { 729 };
   int64_t reshape_186_0_dim_4 { 72 };
   int64_t split_187_0_dim_0 { 1 };
   int64_t split_187_1_dim_0 { 1 };
   int64_t split_187_2_dim_0 { 1 };
   int64_t reshape_188_0_dim_0 { 8 };
   int64_t reshape_188_0_dim_1 { 16 };
   int64_t reshape_188_0_dim_2 { 729 };
   int64_t reshape_188_0_dim_3 { 72 };
   int64_t reshape_189_0_dim_0 { 8 };
   int64_t reshape_189_0_dim_1 { 16 };
   int64_t reshape_189_0_dim_2 { 729 };
   int64_t reshape_189_0_dim_3 { 72 };
   int64_t reshape_190_0_dim_0 { 8 };
   int64_t reshape_190_0_dim_1 { 16 };
   int64_t reshape_190_0_dim_2 { 729 };
   int64_t reshape_190_0_dim_3 { 72 };
   int64_t reshape_194_0_dim_0 { 8 };
   int64_t reshape_194_0_dim_1 { 729 };
   int64_t reshape_194_0_dim_2 { 1152 };
   int64_t reshape_192_0_dim_0 { 5832 };
   int64_t reshape_192_0_dim_1 { 1152 };
   int64_t reshape_199_0_dim_0 { 5832 };
   int64_t reshape_199_0_dim_1 { 1152 };
   int64_t reshape_201_0_dim_0 { 3 };
   int64_t reshape_201_0_dim_1 { 8 };
   int64_t reshape_201_0_dim_2 { 16 };
   int64_t reshape_201_0_dim_3 { 729 };
   int64_t reshape_201_0_dim_4 { 72 };
   int64_t split_202_0_dim_0 { 1 };
   int64_t split_202_1_dim_0 { 1 };
   int64_t split_202_2_dim_0 { 1 };
   int64_t reshape_203_0_dim_0 { 8 };
   int64_t reshape_203_0_dim_1 { 16 };
   int64_t reshape_203_0_dim_2 { 729 };
   int64_t reshape_203_0_dim_3 { 72 };
   int64_t reshape_204_0_dim_0 { 8 };
   int64_t reshape_204_0_dim_1 { 16 };
   int64_t reshape_204_0_dim_2 { 729 };
   int64_t reshape_204_0_dim_3 { 72 };
   int64_t reshape_205_0_dim_0 { 8 };
   int64_t reshape_205_0_dim_1 { 16 };
   int64_t reshape_205_0_dim_2 { 729 };
   int64_t reshape_205_0_dim_3 { 72 };
   int64_t reshape_209_0_dim_0 { 8 };
   int64_t reshape_209_0_dim_1 { 729 };
   int64_t reshape_209_0_dim_2 { 1152 };
   int64_t reshape_207_0_dim_0 { 5832 };
   int64_t reshape_207_0_dim_1 { 1152 };
   int64_t reshape_214_0_dim_0 { 5832 };
   int64_t reshape_214_0_dim_1 { 1152 };
   int64_t reshape_216_0_dim_0 { 3 };
   int64_t reshape_216_0_dim_1 { 8 };
   int64_t reshape_216_0_dim_2 { 16 };
   int64_t reshape_216_0_dim_3 { 729 };
   int64_t reshape_216_0_dim_4 { 72 };
   int64_t split_217_0_dim_0 { 1 };
   int64_t split_217_1_dim_0 { 1 };
   int64_t split_217_2_dim_0 { 1 };
   int64_t reshape_218_0_dim_0 { 8 };
   int64_t reshape_218_0_dim_1 { 16 };
   int64_t reshape_218_0_dim_2 { 729 };
   int64_t reshape_218_0_dim_3 { 72 };
   int64_t reshape_219_0_dim_0 { 8 };
   int64_t reshape_219_0_dim_1 { 16 };
   int64_t reshape_219_0_dim_2 { 729 };
   int64_t reshape_219_0_dim_3 { 72 };
   int64_t reshape_220_0_dim_0 { 8 };
   int64_t reshape_220_0_dim_1 { 16 };
   int64_t reshape_220_0_dim_2 { 729 };
   int64_t reshape_220_0_dim_3 { 72 };
   int64_t reshape_224_0_dim_0 { 8 };
   int64_t reshape_224_0_dim_1 { 729 };
   int64_t reshape_224_0_dim_2 { 1152 };
   int64_t reshape_222_0_dim_0 { 5832 };
   int64_t reshape_222_0_dim_1 { 1152 };
   int64_t reshape_229_0_dim_0 { 5832 };
   int64_t reshape_229_0_dim_1 { 1152 };
   int64_t reshape_231_0_dim_0 { 3 };
   int64_t reshape_231_0_dim_1 { 8 };
   int64_t reshape_231_0_dim_2 { 16 };
   int64_t reshape_231_0_dim_3 { 729 };
   int64_t reshape_231_0_dim_4 { 72 };
   int64_t split_232_0_dim_0 { 1 };
   int64_t split_232_1_dim_0 { 1 };
   int64_t split_232_2_dim_0 { 1 };
   int64_t reshape_233_0_dim_0 { 8 };
   int64_t reshape_233_0_dim_1 { 16 };
   int64_t reshape_233_0_dim_2 { 729 };
   int64_t reshape_233_0_dim_3 { 72 };
   int64_t reshape_234_0_dim_0 { 8 };
   int64_t reshape_234_0_dim_1 { 16 };
   int64_t reshape_234_0_dim_2 { 729 };
   int64_t reshape_234_0_dim_3 { 72 };
   int64_t reshape_235_0_dim_0 { 8 };
   int64_t reshape_235_0_dim_1 { 16 };
   int64_t reshape_235_0_dim_2 { 729 };
   int64_t reshape_235_0_dim_3 { 72 };
   int64_t reshape_239_0_dim_0 { 8 };
   int64_t reshape_239_0_dim_1 { 729 };
   int64_t reshape_239_0_dim_2 { 1152 };
   int64_t reshape_237_0_dim_0 { 5832 };
   int64_t reshape_237_0_dim_1 { 1152 };
   int64_t reshape_244_0_dim_0 { 5832 };
   int64_t reshape_244_0_dim_1 { 1152 };
   int64_t reshape_246_0_dim_0 { 3 };
   int64_t reshape_246_0_dim_1 { 8 };
   int64_t reshape_246_0_dim_2 { 16 };
   int64_t reshape_246_0_dim_3 { 729 };
   int64_t reshape_246_0_dim_4 { 72 };
   int64_t split_247_0_dim_0 { 1 };
   int64_t split_247_1_dim_0 { 1 };
   int64_t split_247_2_dim_0 { 1 };
   int64_t reshape_248_0_dim_0 { 8 };
   int64_t reshape_248_0_dim_1 { 16 };
   int64_t reshape_248_0_dim_2 { 729 };
   int64_t reshape_248_0_dim_3 { 72 };
   int64_t reshape_249_0_dim_0 { 8 };
   int64_t reshape_249_0_dim_1 { 16 };
   int64_t reshape_249_0_dim_2 { 729 };
   int64_t reshape_249_0_dim_3 { 72 };
   int64_t reshape_250_0_dim_0 { 8 };
   int64_t reshape_250_0_dim_1 { 16 };
   int64_t reshape_250_0_dim_2 { 729 };
   int64_t reshape_250_0_dim_3 { 72 };
   int64_t reshape_254_0_dim_0 { 8 };
   int64_t reshape_254_0_dim_1 { 729 };
   int64_t reshape_254_0_dim_2 { 1152 };
   int64_t reshape_252_0_dim_0 { 5832 };
   int64_t reshape_252_0_dim_1 { 1152 };
   int64_t reshape_259_0_dim_0 { 5832 };
   int64_t reshape_259_0_dim_1 { 1152 };
   int64_t reshape_261_0_dim_0 { 3 };
   int64_t reshape_261_0_dim_1 { 8 };
   int64_t reshape_261_0_dim_2 { 16 };
   int64_t reshape_261_0_dim_3 { 729 };
   int64_t reshape_261_0_dim_4 { 72 };
   int64_t split_262_0_dim_0 { 1 };
   int64_t split_262_1_dim_0 { 1 };
   int64_t split_262_2_dim_0 { 1 };
   int64_t reshape_263_0_dim_0 { 8 };
   int64_t reshape_263_0_dim_1 { 16 };
   int64_t reshape_263_0_dim_2 { 729 };
   int64_t reshape_263_0_dim_3 { 72 };
   int64_t reshape_264_0_dim_0 { 8 };
   int64_t reshape_264_0_dim_1 { 16 };
   int64_t reshape_264_0_dim_2 { 729 };
   int64_t reshape_264_0_dim_3 { 72 };
   int64_t reshape_265_0_dim_0 { 8 };
   int64_t reshape_265_0_dim_1 { 16 };
   int64_t reshape_265_0_dim_2 { 729 };
   int64_t reshape_265_0_dim_3 { 72 };
   int64_t reshape_269_0_dim_0 { 8 };
   int64_t reshape_269_0_dim_1 { 729 };
   int64_t reshape_269_0_dim_2 { 1152 };
   int64_t reshape_267_0_dim_0 { 5832 };
   int64_t reshape_267_0_dim_1 { 1152 };
   int64_t reshape_274_0_dim_0 { 5832 };
   int64_t reshape_274_0_dim_1 { 1152 };
   int64_t reshape_276_0_dim_0 { 3 };
   int64_t reshape_276_0_dim_1 { 8 };
   int64_t reshape_276_0_dim_2 { 16 };
   int64_t reshape_276_0_dim_3 { 729 };
   int64_t reshape_276_0_dim_4 { 72 };
   int64_t split_277_0_dim_0 { 1 };
   int64_t split_277_1_dim_0 { 1 };
   int64_t split_277_2_dim_0 { 1 };
   int64_t reshape_278_0_dim_0 { 8 };
   int64_t reshape_278_0_dim_1 { 16 };
   int64_t reshape_278_0_dim_2 { 729 };
   int64_t reshape_278_0_dim_3 { 72 };
   int64_t reshape_279_0_dim_0 { 8 };
   int64_t reshape_279_0_dim_1 { 16 };
   int64_t reshape_279_0_dim_2 { 729 };
   int64_t reshape_279_0_dim_3 { 72 };
   int64_t reshape_280_0_dim_0 { 8 };
   int64_t reshape_280_0_dim_1 { 16 };
   int64_t reshape_280_0_dim_2 { 729 };
   int64_t reshape_280_0_dim_3 { 72 };
   int64_t reshape_284_0_dim_0 { 8 };
   int64_t reshape_284_0_dim_1 { 729 };
   int64_t reshape_284_0_dim_2 { 1152 };
   int64_t reshape_282_0_dim_0 { 5832 };
   int64_t reshape_282_0_dim_1 { 1152 };
   int64_t reshape_289_0_dim_0 { 5832 };
   int64_t reshape_289_0_dim_1 { 1152 };
   int64_t reshape_291_0_dim_0 { 3 };
   int64_t reshape_291_0_dim_1 { 8 };
   int64_t reshape_291_0_dim_2 { 16 };
   int64_t reshape_291_0_dim_3 { 729 };
   int64_t reshape_291_0_dim_4 { 72 };
   int64_t split_292_0_dim_0 { 1 };
   int64_t split_292_1_dim_0 { 1 };
   int64_t split_292_2_dim_0 { 1 };
   int64_t reshape_293_0_dim_0 { 8 };
   int64_t reshape_293_0_dim_1 { 16 };
   int64_t reshape_293_0_dim_2 { 729 };
   int64_t reshape_293_0_dim_3 { 72 };
   int64_t reshape_294_0_dim_0 { 8 };
   int64_t reshape_294_0_dim_1 { 16 };
   int64_t reshape_294_0_dim_2 { 729 };
   int64_t reshape_294_0_dim_3 { 72 };
   int64_t reshape_295_0_dim_0 { 8 };
   int64_t reshape_295_0_dim_1 { 16 };
   int64_t reshape_295_0_dim_2 { 729 };
   int64_t reshape_295_0_dim_3 { 72 };
   int64_t reshape_299_0_dim_0 { 8 };
   int64_t reshape_299_0_dim_1 { 729 };
   int64_t reshape_299_0_dim_2 { 1152 };
   int64_t reshape_297_0_dim_0 { 5832 };
   int64_t reshape_297_0_dim_1 { 1152 };
   int64_t reshape_304_0_dim_0 { 5832 };
   int64_t reshape_304_0_dim_1 { 1152 };
   int64_t reshape_306_0_dim_0 { 3 };
   int64_t reshape_306_0_dim_1 { 8 };
   int64_t reshape_306_0_dim_2 { 16 };
   int64_t reshape_306_0_dim_3 { 729 };
   int64_t reshape_306_0_dim_4 { 72 };
   int64_t split_307_0_dim_0 { 1 };
   int64_t split_307_1_dim_0 { 1 };
   int64_t split_307_2_dim_0 { 1 };
   int64_t reshape_308_0_dim_0 { 8 };
   int64_t reshape_308_0_dim_1 { 16 };
   int64_t reshape_308_0_dim_2 { 729 };
   int64_t reshape_308_0_dim_3 { 72 };
   int64_t reshape_309_0_dim_0 { 8 };
   int64_t reshape_309_0_dim_1 { 16 };
   int64_t reshape_309_0_dim_2 { 729 };
   int64_t reshape_309_0_dim_3 { 72 };
   int64_t reshape_310_0_dim_0 { 8 };
   int64_t reshape_310_0_dim_1 { 16 };
   int64_t reshape_310_0_dim_2 { 729 };
   int64_t reshape_310_0_dim_3 { 72 };
   int64_t reshape_314_0_dim_0 { 8 };
   int64_t reshape_314_0_dim_1 { 729 };
   int64_t reshape_314_0_dim_2 { 1152 };
   int64_t reshape_312_0_dim_0 { 5832 };
   int64_t reshape_312_0_dim_1 { 1152 };
   int64_t reshape_319_0_dim_0 { 5832 };
   int64_t reshape_319_0_dim_1 { 1152 };
   int64_t reshape_321_0_dim_0 { 3 };
   int64_t reshape_321_0_dim_1 { 8 };
   int64_t reshape_321_0_dim_2 { 16 };
   int64_t reshape_321_0_dim_3 { 729 };
   int64_t reshape_321_0_dim_4 { 72 };
   int64_t split_322_0_dim_0 { 1 };
   int64_t split_322_1_dim_0 { 1 };
   int64_t split_322_2_dim_0 { 1 };
   int64_t reshape_323_0_dim_0 { 8 };
   int64_t reshape_323_0_dim_1 { 16 };
   int64_t reshape_323_0_dim_2 { 729 };
   int64_t reshape_323_0_dim_3 { 72 };
   int64_t reshape_324_0_dim_0 { 8 };
   int64_t reshape_324_0_dim_1 { 16 };
   int64_t reshape_324_0_dim_2 { 729 };
   int64_t reshape_324_0_dim_3 { 72 };
   int64_t reshape_325_0_dim_0 { 8 };
   int64_t reshape_325_0_dim_1 { 16 };
   int64_t reshape_325_0_dim_2 { 729 };
   int64_t reshape_325_0_dim_3 { 72 };
   int64_t reshape_329_0_dim_0 { 8 };
   int64_t reshape_329_0_dim_1 { 729 };
   int64_t reshape_329_0_dim_2 { 1152 };
   int64_t reshape_327_0_dim_0 { 5832 };
   int64_t reshape_327_0_dim_1 { 1152 };
   int64_t reshape_334_0_dim_0 { 5832 };
   int64_t reshape_334_0_dim_1 { 1152 };
   int64_t reshape_336_0_dim_0 { 3 };
   int64_t reshape_336_0_dim_1 { 8 };
   int64_t reshape_336_0_dim_2 { 16 };
   int64_t reshape_336_0_dim_3 { 729 };
   int64_t reshape_336_0_dim_4 { 72 };
   int64_t split_337_0_dim_0 { 1 };
   int64_t split_337_1_dim_0 { 1 };
   int64_t split_337_2_dim_0 { 1 };
   int64_t reshape_338_0_dim_0 { 8 };
   int64_t reshape_338_0_dim_1 { 16 };
   int64_t reshape_338_0_dim_2 { 729 };
   int64_t reshape_338_0_dim_3 { 72 };
   int64_t reshape_339_0_dim_0 { 8 };
   int64_t reshape_339_0_dim_1 { 16 };
   int64_t reshape_339_0_dim_2 { 729 };
   int64_t reshape_339_0_dim_3 { 72 };
   int64_t reshape_340_0_dim_0 { 8 };
   int64_t reshape_340_0_dim_1 { 16 };
   int64_t reshape_340_0_dim_2 { 729 };
   int64_t reshape_340_0_dim_3 { 72 };
   int64_t reshape_344_0_dim_0 { 8 };
   int64_t reshape_344_0_dim_1 { 729 };
   int64_t reshape_344_0_dim_2 { 1152 };
   int64_t reshape_342_0_dim_0 { 5832 };
   int64_t reshape_342_0_dim_1 { 1152 };
   int64_t reshape_349_0_dim_0 { 5832 };
   int64_t reshape_349_0_dim_1 { 1152 };
   int64_t reshape_351_0_dim_0 { 3 };
   int64_t reshape_351_0_dim_1 { 8 };
   int64_t reshape_351_0_dim_2 { 16 };
   int64_t reshape_351_0_dim_3 { 729 };
   int64_t reshape_351_0_dim_4 { 72 };
   int64_t split_352_0_dim_0 { 1 };
   int64_t split_352_1_dim_0 { 1 };
   int64_t split_352_2_dim_0 { 1 };
   int64_t reshape_353_0_dim_0 { 8 };
   int64_t reshape_353_0_dim_1 { 16 };
   int64_t reshape_353_0_dim_2 { 729 };
   int64_t reshape_353_0_dim_3 { 72 };
   int64_t reshape_354_0_dim_0 { 8 };
   int64_t reshape_354_0_dim_1 { 16 };
   int64_t reshape_354_0_dim_2 { 729 };
   int64_t reshape_354_0_dim_3 { 72 };
   int64_t reshape_355_0_dim_0 { 8 };
   int64_t reshape_355_0_dim_1 { 16 };
   int64_t reshape_355_0_dim_2 { 729 };
   int64_t reshape_355_0_dim_3 { 72 };
   int64_t reshape_359_0_dim_0 { 8 };
   int64_t reshape_359_0_dim_1 { 729 };
   int64_t reshape_359_0_dim_2 { 1152 };
   int64_t reshape_357_0_dim_0 { 5832 };
   int64_t reshape_357_0_dim_1 { 1152 };
   int64_t reshape_364_0_dim_0 { 5832 };
   int64_t reshape_364_0_dim_1 { 1152 };
   int64_t reshape_366_0_dim_0 { 3 };
   int64_t reshape_366_0_dim_1 { 8 };
   int64_t reshape_366_0_dim_2 { 16 };
   int64_t reshape_366_0_dim_3 { 729 };
   int64_t reshape_366_0_dim_4 { 72 };
   int64_t split_367_0_dim_0 { 1 };
   int64_t split_367_1_dim_0 { 1 };
   int64_t split_367_2_dim_0 { 1 };
   int64_t reshape_368_0_dim_0 { 8 };
   int64_t reshape_368_0_dim_1 { 16 };
   int64_t reshape_368_0_dim_2 { 729 };
   int64_t reshape_368_0_dim_3 { 72 };
   int64_t reshape_369_0_dim_0 { 8 };
   int64_t reshape_369_0_dim_1 { 16 };
   int64_t reshape_369_0_dim_2 { 729 };
   int64_t reshape_369_0_dim_3 { 72 };
   int64_t reshape_370_0_dim_0 { 8 };
   int64_t reshape_370_0_dim_1 { 16 };
   int64_t reshape_370_0_dim_2 { 729 };
   int64_t reshape_370_0_dim_3 { 72 };
   int64_t reshape_374_0_dim_0 { 8 };
   int64_t reshape_374_0_dim_1 { 729 };
   int64_t reshape_374_0_dim_2 { 1152 };
   int64_t reshape_372_0_dim_0 { 5832 };
   int64_t reshape_372_0_dim_1 { 1152 };
   int64_t reshape_379_0_dim_0 { 5832 };
   int64_t reshape_379_0_dim_1 { 1152 };
   int64_t reshape_381_0_dim_0 { 3 };
   int64_t reshape_381_0_dim_1 { 8 };
   int64_t reshape_381_0_dim_2 { 16 };
   int64_t reshape_381_0_dim_3 { 729 };
   int64_t reshape_381_0_dim_4 { 72 };
   int64_t split_382_0_dim_0 { 1 };
   int64_t split_382_1_dim_0 { 1 };
   int64_t split_382_2_dim_0 { 1 };
   int64_t reshape_383_0_dim_0 { 8 };
   int64_t reshape_383_0_dim_1 { 16 };
   int64_t reshape_383_0_dim_2 { 729 };
   int64_t reshape_383_0_dim_3 { 72 };
   int64_t reshape_384_0_dim_0 { 8 };
   int64_t reshape_384_0_dim_1 { 16 };
   int64_t reshape_384_0_dim_2 { 729 };
   int64_t reshape_384_0_dim_3 { 72 };
   int64_t reshape_385_0_dim_0 { 8 };
   int64_t reshape_385_0_dim_1 { 16 };
   int64_t reshape_385_0_dim_2 { 729 };
   int64_t reshape_385_0_dim_3 { 72 };
   int64_t reshape_389_0_dim_0 { 8 };
   int64_t reshape_389_0_dim_1 { 729 };
   int64_t reshape_389_0_dim_2 { 1152 };
   int64_t reshape_387_0_dim_0 { 5832 };
   int64_t reshape_387_0_dim_1 { 1152 };
   int64_t reshape_394_0_dim_0 { 5832 };
   int64_t reshape_394_0_dim_1 { 1152 };
   int64_t reshape_396_0_dim_0 { 3 };
   int64_t reshape_396_0_dim_1 { 8 };
   int64_t reshape_396_0_dim_2 { 16 };
   int64_t reshape_396_0_dim_3 { 729 };
   int64_t reshape_396_0_dim_4 { 72 };
   int64_t split_397_0_dim_0 { 1 };
   int64_t split_397_1_dim_0 { 1 };
   int64_t split_397_2_dim_0 { 1 };
   int64_t reshape_398_0_dim_0 { 8 };
   int64_t reshape_398_0_dim_1 { 16 };
   int64_t reshape_398_0_dim_2 { 729 };
   int64_t reshape_398_0_dim_3 { 72 };
   int64_t reshape_399_0_dim_0 { 8 };
   int64_t reshape_399_0_dim_1 { 16 };
   int64_t reshape_399_0_dim_2 { 729 };
   int64_t reshape_399_0_dim_3 { 72 };
   int64_t reshape_400_0_dim_0 { 8 };
   int64_t reshape_400_0_dim_1 { 16 };
   int64_t reshape_400_0_dim_2 { 729 };
   int64_t reshape_400_0_dim_3 { 72 };
   int64_t reshape_404_0_dim_0 { 8 };
   int64_t reshape_404_0_dim_1 { 729 };
   int64_t reshape_404_0_dim_2 { 1152 };
   int64_t reshape_402_0_dim_0 { 5832 };
   int64_t reshape_402_0_dim_1 { 1152 };
   int64_t reshape_410_0_dim_0 { 8 };
   int64_t reshape_410_0_dim_1 { 729 };
   int64_t reshape_410_0_dim_2 { 2 };
   int64_t reshape_410_0_dim_3 { 16 };
   int64_t reshape_410_0_dim_4 { 72 };
   int64_t split_412_0_dim_0 { 1 };
   int64_t split_412_1_dim_0 { 1 };
   int64_t reshape_417_0_dim_0 { 8 };
   int64_t reshape_417_0_dim_1 { 16 };
   int64_t reshape_417_0_dim_2 { 1 };
   int64_t reshape_417_0_dim_3 { 72 };
   int64_t reshape_420_0_dim_0 { 8 };
   int64_t reshape_420_0_dim_1 { 1 };
   int64_t reshape_420_0_dim_2 { 1152 };


};
} // namespace ait