Upload idm/config.yaml with huggingface_hub

idm/config.yaml

@@ -0,0 +1,363 @@
+output_dir: ./runs/robotwin_idm_3cam_384_1e-4_ep10_clean_only_fewshot/2026-08-10_10-43-33
+batch_size: null
+global_batch_size: 128
+num_workers: 8
+lr_scheduler_type: cosine
+learning_rate: 0.0001
+num_epochs: 10
+max_steps: null
+log_every: 10
+save_every: 5000
+save_state_every: 500000
+eval_every: 500
+eval_num_inference_steps: 10
+gradient_accumulation_steps: 1
+split_video_action_loss_backward: false
+split_video_action_loss_mode: alternating_batches
+mixed_precision: bf16
+seed: 42
+max_grad_norm: 1.0
+weight_decay: 0.01
+resume: null
+sampling:
+  weighted: false
+  default_weight: 1.0
+  combine: max
+  replacement: true
+  num_samples: null
+  rules: []
+overwrite_video_latents: false
+video_latent_save_dtype: fp16
+video_latent_batch_size: 8
+video_latent_num_workers: 8
+wandb:
+  enabled: true
+  workspace: tehaji007
+  project: fast-wam
+  name: robotwin_idm_3cam_384_1e-4_ep10_clean_only_fewshot
+  group: null
+  mode: offline
+data:
+  train:
+    _target_: fastwam.datasets.lerobot.robot_video_dataset.RobotVideoDataset
+    dataset_dirs:
+    - ./data/robotwin2.0/robotwin2.0
+    shape_meta:
+      images:
+      - key: cam_high
+        raw_shape:
+        - 3
+        - 480
+        - 640
+        shape:
+        - 3
+        - 240
+        - 320
+      - key: cam_left_wrist
+        raw_shape:
+        - 3
+        - 480
+        - 640
+        shape:
+        - 3
+        - 240
+        - 320
+      - key: cam_right_wrist
+        raw_shape:
+        - 3
+        - 480
+        - 640
+        shape:
+        - 3
+        - 240
+        - 320
+      action:
+      - key: default
+        raw_shape: 14
+        shape: 14
+      state:
+      - key: default
+        raw_shape: 14
+        shape: 14
+    num_frames: 33
+    global_sample_stride: 1
+    action_video_freq_ratio: 4
+    video_size:
+    - 384
+    - 320
+    camera_key: null
+    val_set_proportion: 0.01
+    is_training_set: true
+    pretrained_norm_stats: ./data/robotwin2.0/dataset_stats.json
+    skip_padding_as_possible: false
+    concat_multi_camera: robotwin
+    use_precomputed_video_latents: true
+    video_latent_cache_dir: ./data/video_latents/robotwin_uncond_3cam_384_1e-4_all_tasks
+    strict_video_latent_loading: true
+    video_latent_storage_format: npz_compressed
+    video_latent_shard_size: 10000
+    use_precomputed_metadata_cache: true
+    metadata_cache_dir: ./data/video_latents/robotwin_uncond_3cam_384_1e-4_all_tasks_metadata_cache
+    metadata_cache_shard_size: 10000
+    subset_sample_indices_file: null
+    prompt_subset_file: null
+    robotwin_clean_only: true
+    robotwin_clean_group_size: 550
+    robotwin_clean_episodes_per_group: 55
+    robotwin_clean_random_select: true
+    robotwin_clean_random_seed: 0
+    robotwin_clean_pool_episodes_per_group: 550
+    processor:
+      _target_: fastwam.datasets.lerobot.processors.fastwam_processor.FastWAMProcessor
+      shape_meta:
+        images:
+        - key: cam_high
+          raw_shape:
+          - 3
+          - 480
+          - 640
+          shape:
+          - 3
+          - 240
+          - 320
+        - key: cam_left_wrist
+          raw_shape:
+          - 3
+          - 480
+          - 640
+          shape:
+          - 3
+          - 240
+          - 320
+        - key: cam_right_wrist
+          raw_shape:
+          - 3
+          - 480
+          - 640
+          shape:
+          - 3
+          - 240
+          - 320
+        action:
+        - key: default
+          raw_shape: 14
+          shape: 14
+        state:
+        - key: default
+          raw_shape: 14
+          shape: 14
+      num_obs_steps: 33
+      num_output_cameras: 3
+      action_output_dim: 14
+      proprio_output_dim: 14
+      action_state_transforms: null
+      use_stepwise_action_norm: false
+      norm_default_mode: z-score
+      norm_exception_mode: null
+      action_state_merger:
+        _target_: fastwam.datasets.lerobot.transforms.action_state_merger.ConcatLeftAlign
+      train_transforms:
+      - _target_: fastwam.datasets.lerobot.transforms.image.ToTensor
+      - _target_: torchvision.transforms.Resize
+        size:
+        - 240
+        - 320
+      val_transforms:
+      - _target_: fastwam.datasets.lerobot.transforms.image.ToTensor
+      - _target_: torchvision.transforms.Resize
+        size:
+        - 240
+        - 320
+    text_embedding_cache_dir: ./data/text_embeds_cache/robotwin
+    context_len: 128
+  val:
+    _target_: fastwam.datasets.lerobot.robot_video_dataset.RobotVideoDataset
+    dataset_dirs:
+    - ./data/robotwin2.0/robotwin2.0/
+    shape_meta:
+      images:
+      - key: cam_high
+        raw_shape:
+        - 3
+        - 480
+        - 640
+        shape:
+        - 3
+        - 240
+        - 320
+      - key: cam_left_wrist
+        raw_shape:
+        - 3
+        - 480
+        - 640
+        shape:
+        - 3
+        - 240
+        - 320
+      - key: cam_right_wrist
+        raw_shape:
+        - 3
+        - 480
+        - 640
+        shape:
+        - 3
+        - 240
+        - 320
+      action:
+      - key: default
+        raw_shape: 14
+        shape: 14
+      state:
+      - key: default
+        raw_shape: 14
+        shape: 14
+    num_frames: 33
+    global_sample_stride: 1
+    action_video_freq_ratio: 4
+    video_size:
+    - 384
+    - 320
+    camera_key: null
+    val_set_proportion: 0.01
+    is_training_set: false
+    pretrained_norm_stats: ./data/robotwin2.0/dataset_stats.json
+    skip_padding_as_possible: false
+    concat_multi_camera: robotwin
+    use_precomputed_video_latents: false
+    video_latent_cache_dir: null
+    strict_video_latent_loading: true
+    video_latent_storage_format: npz_compressed
+    video_latent_shard_size: 10000
+    use_precomputed_metadata_cache: false
+    metadata_cache_dir: null
+    metadata_cache_shard_size: 10000
+    subset_sample_indices_file: null
+    prompt_subset_file: null
+    robotwin_clean_only: true
+    robotwin_clean_group_size: 550
+    robotwin_clean_episodes_per_group: 50
+    robotwin_clean_random_select: false
+    robotwin_clean_random_seed: 0
+    robotwin_clean_pool_episodes_per_group: 50
+    processor:
+      _target_: fastwam.datasets.lerobot.processors.fastwam_processor.FastWAMProcessor
+      shape_meta:
+        images:
+        - key: cam_high
+          raw_shape:
+          - 3
+          - 480
+          - 640
+          shape:
+          - 3
+          - 240
+          - 320
+        - key: cam_left_wrist
+          raw_shape:
+          - 3
+          - 480
+          - 640
+          shape:
+          - 3
+          - 240
+          - 320
+        - key: cam_right_wrist
+          raw_shape:
+          - 3
+          - 480
+          - 640
+          shape:
+          - 3
+          - 240
+          - 320
+        action:
+        - key: default
+          raw_shape: 14
+          shape: 14
+        state:
+        - key: default
+          raw_shape: 14
+          shape: 14
+      num_obs_steps: 33
+      num_output_cameras: 3
+      action_output_dim: 14
+      proprio_output_dim: 14
+      action_state_transforms: null
+      use_stepwise_action_norm: false
+      norm_default_mode: z-score
+      norm_exception_mode: null
+      action_state_merger:
+        _target_: fastwam.datasets.lerobot.transforms.action_state_merger.ConcatLeftAlign
+      train_transforms:
+      - _target_: fastwam.datasets.lerobot.transforms.image.ToTensor
+      - _target_: torchvision.transforms.Resize
+        size:
+        - 240
+        - 320
+      val_transforms:
+      - _target_: fastwam.datasets.lerobot.transforms.image.ToTensor
+      - _target_: torchvision.transforms.Resize
+        size:
+        - 240
+        - 320
+    text_embedding_cache_dir: ./data/text_embeds_cache/robotwin
+    context_len: 128
+model:
+  _target_: fastwam.runtime.create_fastwam_idm
+  model_id: Wan-AI/Wan2.2-TI2V-5B
+  tokenizer_model_id: Wan-AI/Wan2.1-T2V-1.3B
+  tokenizer_max_len: 128
+  load_text_encoder: false
+  proprio_dim: 14
+  redirect_common_files: false
+  mot_checkpoint_mixed_attn: false
+  action_dit_pretrained_path: checkpoints/ActionDiT_linear_interp_Wan22_alphascale_1024hdim.pt
+  skip_dit_load_from_pretrain: false
+  keep_dit_layers: null
+  keep_dit_layer_indices: null
+  video_dit_config:
+    has_image_input: false
+    patch_size:
+    - 1
+    - 2
+    - 2
+    in_dim: 48
+    hidden_dim: 3072
+    ffn_dim: 14336
+    freq_dim: 256
+    text_dim: 4096
+    out_dim: 48
+    num_heads: 24
+    attn_head_dim: 128
+    num_layers: 30
+    eps: 1.0e-06
+    seperated_timestep: true
+    require_clip_embedding: false
+    require_vae_embedding: false
+    fuse_vae_embedding_in_latents: true
+    use_gradient_checkpointing: false
+    video_attention_mask_mode: first_frame_causal
+    action_conditioned: false
+    action_dim: 14
+    action_group_causal_mask_mode: group_diagonal
+  action_dit_config:
+    action_dim: 14
+    hidden_dim: 1024
+    ffn_dim: 4096
+    num_heads: 24
+    attn_head_dim: 128
+    num_layers: 30
+    text_dim: 4096
+    freq_dim: 256
+    eps: 1.0e-06
+    use_gradient_checkpointing: false
+  video_scheduler:
+    train_shift: 5.0
+    infer_shift: 5.0
+    num_train_timesteps: 1000
+  action_scheduler:
+    train_shift: 5.0
+    infer_shift: 5.0
+    num_train_timesteps: 1000
+  loss:
+    lambda_action: 1.0