Instructions to use RLinf/RLinf-DreamZero-WAN2.2-5B-LIBERO-SFT-Diffusers with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use RLinf/RLinf-DreamZero-WAN2.2-5B-LIBERO-SFT-Diffusers with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("RLinf/RLinf-DreamZero-WAN2.2-5B-LIBERO-SFT-Diffusers", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
| { | |
| "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": 16, | |
| "action_loss_embodiment_ids": [ | |
| 26, | |
| 17 | |
| ], | |
| "add_pos_embed": true, | |
| "backbone_embedding_dim": 0, | |
| "backbone_features_projector_cfg": null, | |
| "decouple_inference_noise": false, | |
| "decouple_video_action_noise": false, | |
| "defer_lora_injection": false, | |
| "diffusion_model_cfg": { | |
| "_convert_": "object", | |
| "_target_": "groot.vla.model.dreamzero.modules.wan_video_dit_action_casual_chunk.CausalWanModel", | |
| "concat_first_frame_latent": false, | |
| "diffusion_model_pretrained_path": "/mnt/project_rlinf/hf_models/Wan2.2-TI2V-5B", | |
| "dim": 3072, | |
| "eps": 1e-06, | |
| "ffn_dim": 14336, | |
| "frame_seqlen": 50, | |
| "freq_dim": 256, | |
| "in_dim": 48, | |
| "max_chunk_size": 4, | |
| "model_type": "ti2v", | |
| "num_action_per_block": 16, | |
| "num_frame_per_block": 2, | |
| "num_heads": 24, | |
| "num_layers": 30, | |
| "num_state_per_block": 1, | |
| "out_dim": 48 | |
| }, | |
| "expand_batch": null, | |
| "freeze_decode_layer": false, | |
| "hidden_size": 64, | |
| "high_noise_beta_alpha": 3.0, | |
| "image_encoder_cfg": { | |
| "_convert_": "object", | |
| "_target_": "groot.vla.model.dreamzero.modules.wan_video_image_encoder.WanImageEncoder", | |
| "image_encoder_pretrained_path": "/mnt/project_rlinf_hs/yuanhuining/models/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": 4, | |
| "lora_rank": 4, | |
| "lora_target_modules": "q,k,v,o,ffn.0,ffn.2", | |
| "max_action_dim": 32, | |
| "max_state_dim": 64, | |
| "model_dtype": "bfloat16", | |
| "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, | |
| "target_video_height": 160, | |
| "target_video_width": 320, | |
| "text_encoder_cfg": { | |
| "_convert_": "object", | |
| "_target_": "groot.vla.model.dreamzero.modules.wan_video_text_encoder.WanTextEncoder", | |
| "text_encoder_pretrained_path": "/mnt/project_rlinf/hf_models/Wan2.2-TI2V-5B/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": "full", | |
| "tune_diffusion_model": true, | |
| "tune_projector": true, | |
| "use_gradient_checkpointing": true, | |
| "use_high_noise_emphasis": false, | |
| "use_vlln": true, | |
| "vae_cfg": { | |
| "_convert_": "object", | |
| "_target_": "groot.vla.model.dreamzero.modules.wan_video_vae.WanVideoVAE38", | |
| "dim": 160, | |
| "vae_pretrained_path": "/mnt/project_rlinf/hf_models/Wan2.2-TI2V-5B/Wan2.2_VAE.pth", | |
| "z_dim": 48 | |
| }, | |
| "video_inference_final_noise": 0.8, | |
| "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": 16, | |
| "architectures": [ | |
| "VLA" | |
| ], | |
| "backbone_cfg": { | |
| "_target_": "groot.vla.model.dreamzero.backbone.identity.IdentityBackbone" | |
| }, | |
| "diffusion_model_pretrained_path": "/mnt/project_rlinf/hf_models/Wan2.2-TI2V-5B", | |
| "dtype": "bfloat16", | |
| "embodiment_tag": "libero_sim", | |
| "image_encoder_pretrained_path": "/mnt/project_rlinf_hs/yuanhuining/models/Wan2.1-I2V-14B-480P/models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth", | |
| "is_lora": false, | |
| "max_action_dim": 32, | |
| "max_state_dim": 64, | |
| "metadata_json_path": "/mnt/project_rlinf/guozhen/test_wam/test_dataset/metadata.json", | |
| "model_path": null, | |
| "model_type": "dreamzero", | |
| "num_action_per_block": 16, | |
| "precision": "fp32", | |
| "relative_action": false, | |
| "relative_action_keys": [], | |
| "relative_action_per_horizon": false, | |
| "state_horizon": 1, | |
| "target_video_height": 160, | |
| "target_video_width": 320, | |
| "text_encoder_pretrained_path": "/mnt/project_rlinf/hf_models/Wan2.2-TI2V-5B/models_t5_umt5-xxl-enc-bf16.pth", | |
| "tokenizer_path": "/mnt/project_rlinf/liyanghao/data/models/umt5-xxl", | |
| "transformers_version": "4.57.6", | |
| "vae_pretrained_path": "/mnt/project_rlinf/hf_models/Wan2.2-TI2V-5B/Wan2.2_VAE.pth" | |
| } | |