# Copyright (c) 2023 Graphcore Ltd. All rights reserved.
import numpy as np
import torch

# HF
from transformers.models.gpt_neox.configuration_gpt_neox import GPTNeoXConfig as HFConfig
from transformers.models.gpt_neox.modeling_gpt_neox import GPTNeoXModel

import popxl
from popxl.utils import to_numpy

import popxl_addons as addons
from popxl_addons.patterns import apply_pre_alias_patterns

from config import DollyConfig
from modelling.embedding import DollyEmbeddingsTP
from modelling.dolly_model import DollyModelTP


def test_model_TP_cmp_huggingface(test_config: DollyConfig):
    torch.manual_seed(42)

    batch_size = test_config.execution.micro_batch_size
    hidden_size = test_config.model.hidden_size
    seq_len = test_config.model.sequence_length
    intermediate_size = hidden_size * 4
    # HuggingFace
    config = HFConfig(
        num_hidden_layers=test_config.model.layers,
        vocab_size=test_config.model.embedding.vocab_size,
        hidden_size=hidden_size,
        max_position_embeddings=seq_len,
        intermediate_size=intermediate_size,
        num_attention_heads=test_config.model.attention.heads,
        rotary_dim=test_config.model.attention.rotary_dim,
    )
    hf_model = GPTNeoXModel(config).eval()

    # HF forward
    input_t = torch.randint(0, test_config.model.embedding.vocab_size, (batch_size, test_config.model.sequence_length))
    output_HF = hf_model(input_ids=input_t)[0]
    output_HF = output_HF.detach().numpy()

    # TP
    tp = test_config.execution.tensor_parallel

    # Offset inputs
    words_offsetted = DollyEmbeddingsTP.offset_inputs(test_config, to_numpy(input_t))

    # popxl
    ir = popxl.Ir()
    ir.replication_factor = tp
    replica_grouping = ir.replica_grouping(stride=1, group_size=1)
    main = ir.main_graph

    with main:
        inputs_data, inputs_host_steam, inputs_tensors = zip(
            *[
                addons.host_load(words_offsetted[0], popxl.int32, name="words"),
            ]
        )
        (words,) = inputs_tensors
        facts, graph = DollyModelTP(test_config).create_graph(words)

        vars = facts.init()
        gpt = graph.bind(vars)
        call_info = gpt.call_with_info(words)
        act, *_ = call_info.outputs
        act_stream = addons.host_store(act)

    apply_pre_alias_patterns(ir, level="default")

    # Map weights from huggingface
    weights = DollyModelTP.hf_mapping(test_config, vars, hf_model)

    inputs = dict(zip(inputs_host_steam, [words_offsetted]))

    ir.num_host_transfers = test_config.execution.device_iterations

    with popxl.Session(ir, "ipu_hw") as session:
        session.write_variables_data(weights)
        outs = session.run(inputs)

    # Fwd output
    fwd_data = outs[act_stream]

    assert len(fwd_data) == tp
    for i in range(1, tp):
        np.testing.assert_equal(fwd_data[0], fwd_data[i])

    np.testing.assert_almost_equal(output_HF, fwd_data[0].reshape(output_HF.shape), 3)