lam/config.json

{
  "action_depth": 5,
  "action_dropout": 0.0,
  "action_hidden_dim": 96,
  "action_obs_dim": 0,
  "action_prev_dim": 10,
  "action_state_dim": 5,
  "action_target_dim": 10,
  "action_wide_dim": 512,
  "architectures": [
    "LAMModel"
  ],
  "decoder_attention_head_dim": 64,
  "decoder_attn_implementation": "flash_attention_2",
  "decoder_encoder_hidden_dim": 5,
  "decoder_eps": 1e-06,
  "decoder_ffn_dim": 768,
  "decoder_freq_dim": 64,
  "decoder_in_channels": 3,
  "decoder_num_attention_heads": 3,
  "decoder_num_layers": 12,
  "decoder_out_channels": 3,
  "decoder_patch_size": [
    4,
    4
  ],
  "decoder_pos_embed_seq_len": null,
  "decoder_rope_max_seq_len": 1024,
  "dtype": "bfloat16",
  "encoder_height": 64,
  "encoder_width": 64,
  "fsq_levels": [
    7,
    5,
    5,
    5,
    5
  ],
  "initializer_range": 0.02,
  "is_action_discrete": false,
  "is_diffusion": true,
  "latent_channels": 5,
  "max_tokens": 128,
  "min_tokens": 1,
  "model_type": "lam",
  "null_latent": 0,
  "transformers_version": "4.57.1",
  "use_tail_drop": true,
  "videomae_config": {
    "attn_drop_rate": 0.0,
    "cos_attn": false,
    "depth": 8,
    "drop_path_rate": 0.0,
    "drop_rate": 0.0,
    "embed_dim": 192,
    "img_size": [
      64,
      64
    ],
    "in_chans": 3,
    "init_values": 0.0,
    "layer_norm_eps": 1e-06,
    "mlp_ratio": 4,
    "norm_layer": "nn.LayerNorm",
    "num_classes": 0,
    "num_frames": 2,
    "num_heads": 3,
    "patch_size": 4,
    "qk_scale": null,
    "qkv_bias": true,
    "tubelet_size": 2,
    "use_learnable_pos_emb": false,
    "use_mean_pooling": false,
    "with_cp": false
  },
  "videomae_from_pretrained": null,
  "vocab_size": 4375
}