config.json

{
  "action_dim": 7,
  "architectures": [
    "STDiT3"
  ],
  "caption_channels": 4096,
  "cfg": false,
  "class_dropout_prob": 0.1,
  "depth": 28,
  "drop_path": 0.0,
  "enable_flash_attn": true,
  "enable_layernorm_kernel": true,
  "enable_sequence_parallelism": false,
  "freeze_y_embedder": true,
  "hidden_size": 1152,
  "in_channels": 4,
  "input_size": [
    12,
    32,
    32
  ],
  "input_sq_size": 512,
  "mlp_ratio": 4.0,
  "model_max_length": 300,
  "model_type": "STDiT3",
  "num_heads": 16,
  "only_train_temporal": false,
  "pad_action_num": 4,
  "patch_size": [
    1,
    2,
    2
  ],
  "pred_sigma": true,
  "qk_norm": true,
  "skip_y_embedder": true,
  "torch_dtype": "bfloat16",
  "training": true,
  "transformers_version": "4.40.1"
}