config.json

{
  "dtype": "bfloat16",
  "dit_num_layers": 18,
  "dit_num_heads": 8,
  "dit_head_dim": 128,
  "dit_dropout": 0.0,
  "dit_cross_attention_dim": null,
  "dit_interleave_self_attention": true,
  "dit_layerwise_vlm_features": false,
  "chunk_size": 50,
  "n_action_steps": 50,
  "max_state_dim": 32,
  "max_action_dim": 32,
  "num_inference_steps": 10,
  "image_resolution": [
    224,
    224
  ],
  "attn_implementation": "flash_attention_2",
  "tokenizer_max_length": 48,
  "training_phase": "posttrain",
  "action_mode": "delta",
  "knowledge_isolation": true,
  "use_fast_tokenizer": true,
  "discrete_action_vocab_size": 2048
}