| { |
| "action_dim": 32, |
| "action_head_cfg": { |
| "_convert_": "object", |
| "_target_": "groot.vla.model.dreamzero.action_head.wan_flow_matching_action_tf.WANPolicyHead", |
| "config": { |
| "_recursive_": false, |
| "_target_": "groot.vla.model.dreamzero.action_head.wan_flow_matching_action_tf.WANPolicyHeadConfig", |
| "action_dim": 32, |
| "action_horizon": 24, |
| "action_loss_embodiment_ids": [ |
| 26, |
| 17, |
| 32 |
| ], |
| "add_pos_embed": true, |
| "backbone_embedding_dim": 0, |
| "backbone_features_projector_cfg": null, |
| "decouple_video_action_noise": false, |
| "defer_lora_injection": true, |
| "diffusion_model_cfg": { |
| "_convert_": "object", |
| "_target_": "groot.vla.model.dreamzero.modules.wan_video_dit_action_casual_chunk.CausalWanModel", |
| "diffusion_model_pretrained_path": "/mnt/nas3/yunze/research/wam/dreamzero/checkpoints/Wan2.1-I2V-14B-480P", |
| "dim": 5120, |
| "eps": 1e-06, |
| "ffn_dim": 13824, |
| "frame_seqlen": 880, |
| "freq_dim": 256, |
| "in_dim": 36, |
| "max_chunk_size": 4, |
| "model_type": "i2v", |
| "num_action_per_block": 24, |
| "num_frame_per_block": 2, |
| "num_heads": 40, |
| "num_layers": 40, |
| "num_state_per_block": 1, |
| "out_dim": 16 |
| }, |
| "expand_batch": null, |
| "freeze_decode_layer": false, |
| "hidden_size": 64, |
| "image_encoder_cfg": { |
| "_convert_": "object", |
| "_target_": "groot.vla.model.dreamzero.modules.wan_video_image_encoder.WanImageEncoder", |
| "image_encoder_pretrained_path": "/mnt/nas3/yunze/research/wam/dreamzero/checkpoints/Wan2.1-I2V-14B-480P/models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth" |
| }, |
| "init_lora_weights": "kaiming", |
| "input_embedding_dim": 1536, |
| "load_pretrained_det_decode_layer_path": null, |
| "lora_alpha": 16, |
| "lora_rank": 16, |
| "lora_target_modules": "q,k,v,o,ffn.0,ffn.2", |
| "max_action_dim": 32, |
| "max_state_dim": 64, |
| "model_dtype": "float32", |
| "noise_beta_alpha": 1.5, |
| "noise_beta_beta": 1.0, |
| "noise_s": 0.999, |
| "num_frame_per_block": 2, |
| "num_frames": 33, |
| "num_inference_timesteps": 4, |
| "num_timestep_buckets": 1000, |
| "repa_coeff": 1.0, |
| "repa_layer": 8, |
| "skip_component_loading": true, |
| "text_encoder_cfg": { |
| "_convert_": "object", |
| "_target_": "groot.vla.model.dreamzero.modules.wan_video_text_encoder.WanTextEncoder", |
| "text_encoder_pretrained_path": "/mnt/nas3/yunze/research/wam/dreamzero/checkpoints/Wan2.1-I2V-14B-480P/models_t5_umt5-xxl-enc-bf16.pth" |
| }, |
| "tile_size_height": 34, |
| "tile_size_width": 34, |
| "tile_stride_height": 18, |
| "tile_stride_width": 16, |
| "tiled": false, |
| "train_architecture": "lora", |
| "tune_diffusion_model": true, |
| "tune_projector": true, |
| "use_gradient_checkpointing": true, |
| "use_vlln": true, |
| "vae_cfg": { |
| "_convert_": "object", |
| "_target_": "groot.vla.model.dreamzero.modules.wan_video_vae.WanVideoVAE", |
| "vae_pretrained_path": "/mnt/nas3/yunze/research/wam/dreamzero/checkpoints/Wan2.1-I2V-14B-480P/Wan2.1_VAE.pth" |
| }, |
| "video_noise_beta_alpha": 3.0, |
| "video_noise_beta_beta": 1.0, |
| "vl_self_attention_cfg": { |
| "_target_": "groot.vla.model.n1_5.modules.cross_attention_dit.SelfAttentionTransformer", |
| "attention_head_dim": 64, |
| "dropout": 0.2, |
| "final_dropout": true, |
| "num_attention_heads": 24, |
| "num_layers": 4, |
| "positional_embeddings": null |
| } |
| } |
| }, |
| "action_horizon": 24, |
| "architectures": [ |
| "VLA" |
| ], |
| "backbone_cfg": { |
| "_target_": "groot.vla.model.dreamzero.backbone.identity.IdentityBackbone" |
| }, |
| "hidden_size": 0, |
| "model_dtype": "float32", |
| "model_type": "vla", |
| "resume_path": "/mnt/nas3/yunze/research/wam/robodojo/XPolicyLab/policy/DreamZero/checkpoints/RoboDojo-cotrain-arx_x5-3500-joint-42", |
| "torch_dtype": "bfloat16", |
| "transformers_version": "4.51.3" |
| } |
|
|
