config.json

{
  "ONLY_RL_TASKS": true,
  "_name_or_path": "checkpoints/jat-regent-medium-10.0lamda-1.0MDM-1.0ADM-p95DN-resnet18_512ADT_embeddings/checkpoint-27726",
  "action_loss_coef": 1.0,
  "action_vocab_size": 18,
  "activation_function": "gelu_new",
  "architectures": [
    "JatRegentModel"
  ],
  "atari_dist_multiplier": 1.0,
  "atari_dist_type": "resnet18_512",
  "attention_dropout": 0.0,
  "attention_layers": [
    "global",
    "local",
    "global",
    "local",
    "global",
    "local",
    "global",
    "local",
    "global",
    "local",
    "global",
    "local"
  ],
  "attention_types": [
    [
      [
        "global",
        "local"
      ],
      6
    ]
  ],
  "auto_map": {
    "AutoConfig": "configuration_jat.JatConfig",
    "AutoModelForCausalLM": "modeling_jat_regent.JatRegentModel"
  },
  "bos_token_id": 50256,
  "classifier_dropout": 0.1,
  "dist_normalizer": "p95",
  "embed_dropout": 0.0,
  "eos_token_id": 50256,
  "finetune_num_demos": null,
  "hidden_size": 768,
  "image_size": 224,
  "initializer_range": 0.02,
  "intermediate_size": null,
  "lamda": 10.0,
  "layer_norm_epsilon": 1e-05,
  "max_continuous_size": 513,
  "max_discrete_value": 212,
  "max_position_embeddings": 40,
  "model_type": "jat",
  "mujoco_dist_multiplier": 1.0,
  "num_channels": 3,
  "num_contexts": 20,
  "num_heads": 12,
  "num_layers": 12,
  "observation_loss_coef": 0.0,
  "patch_size": 16,
  "resid_dropout": 0.0,
  "tokenizer_class": "GPT2TokenizerFast",
  "torch_dtype": "float32",
  "transformers_version": "4.41.2",
  "use_atari_embeddings": true,
  "use_cache": true,
  "use_global_atari_actions": true,
  "vocab_size": 50257,
  "window_size": 256
}