{
    "output_dir": "/tmp/perceiver-io/",
    "remove_unused_columns": false,
    "do_eval": false,
    "save_strategy": "epoch",
    "logging_strategy": "epoch",

    "pod_type": "pod4",
    "num_train_epochs": 1,
    "per_device_train_batch_size": 1,
    "gradient_accumulation_steps": 51,
    "ipu_config_name": "configs/debug_ipu_config.json",

    "dataset_name": "cifar10",
    "dataset_path": "/localdata/datasets/imagenet_object_localization_patched2019.tar.gz",
    "train_split": 0.03,
    "dataloader_drop_last": true,
    "dataloader_num_workers": 64,
    "dataloader_mode": "async_rebatched",

    "model_name": "deepmind/vision-perceiver-conv",
    "recomputation": true,

    "num_blocks": 2,
    "d_latents": 256,
    "num_latents": 128,
    "num_self_attention_heads": 4,
    "num_cross_attention_heads": 1,
    "num_self_attends_per_block": 3,

    "warmup_ratio": 0.5,
    "learning_rate": 5e-4,
    "constant_cosine": true,

    "lamb": true,
    "adam_epsilon": 1e-6
}