Upload folder using huggingface_hub

dreamzero_libero_goal_full_finetune_w_cache_bs64/checkpoints/step-008270-epoch-10-loss=0.0503.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f144dc7e149a165e9a8ee6b1ac4ad1611e4ae604bc848a090fd5870f0cc9489a
+size 91696959464

dreamzero_libero_goal_full_finetune_w_cache_bs64/config.json

@@ -0,0 +1,291 @@
+{
+  "_ckpt_root": "./checkpoints",
+  "_frame_window_size": 17,
+  "_tokenizer": "./checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl",
+  "eval": {
+    "dataset": {
+      "img_buffer_len": 4,
+      "transforms": [
+        {
+          "img_keys": [
+            "agentview_image",
+            "robot0_eye_in_hand_image"
+          ],
+          "type": "ProcessLiberoEvalInputs"
+        },
+        {
+          "image_resize_strategy": "resize-naive",
+          "input_sizes": [
+            [
+              3,
+              128,
+              128
+            ],
+            [
+              3,
+              128,
+              128
+            ]
+          ],
+          "means": [
+            [
+              127.5,
+              127.5,
+              127.5
+            ],
+            [
+              127.5,
+              127.5,
+              127.5
+            ]
+          ],
+          "stds": [
+            [
+              127.5,
+              127.5,
+              127.5
+            ],
+            [
+              127.5,
+              127.5,
+              127.5
+            ]
+          ],
+          "type": "TransformImage"
+        },
+        {
+          "gripper_key": "robot0_gripper_qpos",
+          "norm_type": "mean_std",
+          "out_key": "states",
+          "pos_key": "robot0_eef_pos",
+          "quat_key": "robot0_eef_quat",
+          "state_dim": 32,
+          "type": "LiberoProprioFromInputs"
+        },
+        {
+          "max_len": 512,
+          "negative_prompt": "Vibrant colors, overexposed, static, blurry details, text, subtitles, style, artwork, painting, image, still, grayscale, dull, worst quality, low quality, JPEG artifacts, ugly, mutilated, extra fingers, bad hands, bad face, deformed, disfigured, mutated limbs, fused fingers, stagnant image, cluttered background, three legs, many people in the background, walking backwards.",
+          "tokenizer": {
+            "model_path": "./checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl",
+            "type": "PretrainedTokenizer"
+          },
+          "type": "LiberoPromptFromInputs",
+          "use_conversation": false
+        },
+        {
+          "frame_window_size": 1,
+          "num_views": 2,
+          "type": "PrepareVideo"
+        }
+      ],
+      "type": "LiberoParquetEvalDataset"
+    },
+    "denormalize_action": {
+      "action_dim": 7,
+      "norm_type": "mean_std",
+      "type": "DenormalizeLiberoAction"
+    },
+    "enable_mixed_precision_training": true,
+    "eval_chunk_size": 10,
+    "mixed_precision_dtype": "bf16",
+    "model_family": "dreamzero",
+    "num_steps_wait": 10,
+    "num_trials_per_task": 50,
+    "resize_size": 128,
+    "seed": 7,
+    "task_suite_name": "libero_goal",
+    "type": "LiberoEvalRunner"
+  },
+  "model": {
+    "frame_window_size": 17,
+    "name_mapping": {
+      "vla_head.model": "action_head.model",
+      "vlm_backbone.image_encoder": "action_head.image_encoder",
+      "vlm_backbone.text_encoder": "action_head.text_encoder",
+      "vlm_backbone.vae": "action_head.vae"
+    },
+    "num_views": 2,
+    "pretrained_name_or_path": "./checkpoints/DreamZero-AgiBot",
+    "type": "DreamZeroVLA",
+    "use_cache": true,
+    "vla_head": {
+      "action_dim": 7,
+      "action_horizon": 10,
+      "cfg_scale": 5.0,
+      "decouple_inference_noise": false,
+      "decouple_video_action_noise": false,
+      "dit_dim": 5120,
+      "dit_ffn_dim": 13824,
+      "dit_freq_dim": 256,
+      "dit_in_dim": 36,
+      "dit_num_heads": 40,
+      "dit_num_layers": 40,
+      "dit_out_dim": 16,
+      "frame_seqlen": 128,
+      "hidden_size": 1024,
+      "input_embedding_dim": 1536,
+      "max_action_dim": 32,
+      "max_chunk_size": 2,
+      "max_num_embodiments": 32,
+      "max_state_dim": 64,
+      "noise_beta_alpha": 1.5,
+      "noise_beta_beta": 1.0,
+      "noise_s": 0.999,
+      "num_action_per_block": 10,
+      "num_frame_per_block": 2,
+      "num_frames": 17,
+      "num_inference_steps": 16,
+      "num_state_per_block": 1,
+      "type": "DreamZeroHead",
+      "use_gradient_checkpointing": true,
+      "video_inference_final_noise": 0.8,
+      "video_noise_beta_alpha": 3.0,
+      "video_noise_beta_beta": 1.0
+    },
+    "vlm_backbone": {
+      "image_encoder_path": null,
+      "text_encoder_path": null,
+      "tiled": false,
+      "type": "WanBackbone",
+      "vae_path": null
+    }
+  },
+  "per_device_num_workers": 4,
+  "runner": {
+    "change_key_name": false,
+    "collator": {
+      "keys": [
+        "states",
+        "images",
+        "img_masks",
+        "actions",
+        "action_masks",
+        "embodiment_ids",
+        "frame_masks",
+        "lang_tokens",
+        "lang_masks"
+      ],
+      "meta_keys": [
+        "task_description",
+        "prompt",
+        "info",
+        "stats",
+        "timestamp"
+      ],
+      "type": "DictCollator"
+    },
+    "enable_gradient_checkpointing": true,
+    "enable_mixed_precision_training": true,
+    "learning_rate": 1e-05,
+    "lr_scheduler_type": "linear-warmup+cosine-decay",
+    "max_epochs": 10,
+    "max_grad_norm": 1.0,
+    "max_keep_ckpts": 1,
+    "metric": {
+      "active_trackers": [
+        "jsonl",
+        "wandb"
+      ],
+      "grad_accumulation_steps": 1,
+      "run_dir": "work_dirs",
+      "type": "VLAMetric",
+      "window_size": 1
+    },
+    "mixed_precision_dtype": "bf16",
+    "sampler": null,
+    "sharding_strategy": "full-shard",
+    "type": "FSDPTrainRunner",
+    "warmup_ratio": 0.05,
+    "weight_decay": 1e-05
+  },
+  "train_dataloader": {
+    "dataset": {
+      "datasets": {
+        "action_key": "action",
+        "action_window_size": 20,
+        "data_root_path": "./datasets/libero_goal_no_noops_lerobotv2.1",
+        "frame_window_size": 17,
+        "statistic_name": "libero_goal_no_noops",
+        "transforms": [
+          {
+            "embodiment_id": 0,
+            "name_mappings": {
+              "actions": [
+                "actions"
+              ],
+              "observation.state": [
+                "states"
+              ]
+            },
+            "parquet_keys": [
+              "observation.state",
+              "timestamp",
+              "actions",
+              "info",
+              "stats",
+              "action_masks"
+            ],
+            "type": "ProcessParquetInputs",
+            "video_keys": [
+              "observation.images.image",
+              "observation.images.wrist_image"
+            ]
+          },
+          {
+            "type": "ParquetPrompter",
+            "use_conversation": false
+          },
+          {
+            "max_len": 512,
+            "tokenizer": {
+              "model_path": "./checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl",
+              "type": "PretrainedTokenizer"
+            },
+            "type": "ProcessPrompts"
+          },
+          {
+            "height": 128,
+            "type": "ResizeImages",
+            "width": 128
+          },
+          {
+            "type": "SimpleNormalizeImages"
+          },
+          {
+            "action_dim": 32,
+            "action_key": "action",
+            "norm_type": "mean_std",
+            "state_dim": 32,
+            "state_key": "proprio",
+            "type": "NormalizeStatesAndActions"
+          },
+          {
+            "frame_window_size": 17,
+            "num_views": 2,
+            "type": "PrepareVideo"
+          }
+        ],
+        "type": "ParquetDataset",
+        "use_delta": false,
+        "window_start_idx": 0
+      },
+      "name_mappings": {
+        "action": [
+          "action"
+        ],
+        "observation.state": [
+          "proprio"
+        ]
+      },
+      "statistic_keys": [
+        "observation.state",
+        "timestamp",
+        "action"
+      ],
+      "statistic_name": "libero_goal_no_noops",
+      "type": "DistributedRepeatingDataset"
+    },
+    "per_device_batch_size": 4,
+    "per_device_num_workers": 4
+  }
+}

dreamzero_libero_goal_full_finetune_w_cache_bs64/config.yaml

@@ -0,0 +1,225 @@
+_ckpt_root: ./checkpoints
+_frame_window_size: 17
+_tokenizer: ./checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl
+eval:
+  dataset:
+    img_buffer_len: 4
+    transforms:
+    - img_keys:
+      - agentview_image
+      - robot0_eye_in_hand_image
+      type: ProcessLiberoEvalInputs
+    - image_resize_strategy: resize-naive
+      input_sizes:
+      - - 3
+        - 128
+        - 128
+      - - 3
+        - 128
+        - 128
+      means:
+      - - 127.5
+        - 127.5
+        - 127.5
+      - - 127.5
+        - 127.5
+        - 127.5
+      stds:
+      - - 127.5
+        - 127.5
+        - 127.5
+      - - 127.5
+        - 127.5
+        - 127.5
+      type: TransformImage
+    - gripper_key: robot0_gripper_qpos
+      norm_type: mean_std
+      out_key: states
+      pos_key: robot0_eef_pos
+      quat_key: robot0_eef_quat
+      state_dim: 32
+      type: LiberoProprioFromInputs
+    - max_len: 512
+      negative_prompt: Vibrant colors, overexposed, static, blurry details, text,
+        subtitles, style, artwork, painting, image, still, grayscale, dull, worst
+        quality, low quality, JPEG artifacts, ugly, mutilated, extra fingers, bad
+        hands, bad face, deformed, disfigured, mutated limbs, fused fingers, stagnant
+        image, cluttered background, three legs, many people in the background, walking
+        backwards.
+      tokenizer:
+        model_path: ./checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl
+        type: PretrainedTokenizer
+      type: LiberoPromptFromInputs
+      use_conversation: false
+    - frame_window_size: 1
+      num_views: 2
+      type: PrepareVideo
+    type: LiberoParquetEvalDataset
+  denormalize_action:
+    action_dim: 7
+    norm_type: mean_std
+    type: DenormalizeLiberoAction
+  enable_mixed_precision_training: true
+  eval_chunk_size: 10
+  mixed_precision_dtype: bf16
+  model_family: dreamzero
+  num_steps_wait: 10
+  num_trials_per_task: 50
+  resize_size: 128
+  seed: 7
+  task_suite_name: libero_goal
+  type: LiberoEvalRunner
+model:
+  frame_window_size: 17
+  name_mapping:
+    vla_head.model: action_head.model
+    vlm_backbone.image_encoder: action_head.image_encoder
+    vlm_backbone.text_encoder: action_head.text_encoder
+    vlm_backbone.vae: action_head.vae
+  num_views: 2
+  pretrained_name_or_path: ./checkpoints/DreamZero-AgiBot
+  type: DreamZeroVLA
+  use_cache: true
+  vla_head:
+    action_dim: 7
+    action_horizon: 10
+    cfg_scale: 5.0
+    decouple_inference_noise: false
+    decouple_video_action_noise: false
+    dit_dim: 5120
+    dit_ffn_dim: 13824
+    dit_freq_dim: 256
+    dit_in_dim: 36
+    dit_num_heads: 40
+    dit_num_layers: 40
+    dit_out_dim: 16
+    frame_seqlen: 128
+    hidden_size: 1024
+    input_embedding_dim: 1536
+    max_action_dim: 32
+    max_chunk_size: 2
+    max_num_embodiments: 32
+    max_state_dim: 64
+    noise_beta_alpha: 1.5
+    noise_beta_beta: 1.0
+    noise_s: 0.999
+    num_action_per_block: 10
+    num_frame_per_block: 2
+    num_frames: 17
+    num_inference_steps: 16
+    num_state_per_block: 1
+    type: DreamZeroHead
+    use_gradient_checkpointing: true
+    video_inference_final_noise: 0.8
+    video_noise_beta_alpha: 3.0
+    video_noise_beta_beta: 1.0
+  vlm_backbone:
+    image_encoder_path: null
+    text_encoder_path: null
+    tiled: false
+    type: WanBackbone
+    vae_path: null
+per_device_num_workers: 4
+runner:
+  change_key_name: false
+  collator:
+    keys:
+    - states
+    - images
+    - img_masks
+    - actions
+    - action_masks
+    - embodiment_ids
+    - frame_masks
+    - lang_tokens
+    - lang_masks
+    meta_keys:
+    - task_description
+    - prompt
+    - info
+    - stats
+    - timestamp
+    type: DictCollator
+  enable_gradient_checkpointing: true
+  enable_mixed_precision_training: true
+  learning_rate: 1.0e-05
+  lr_scheduler_type: linear-warmup+cosine-decay
+  max_epochs: 10
+  max_grad_norm: 1.0
+  max_keep_ckpts: 1
+  metric:
+    active_trackers:
+    - jsonl
+    - wandb
+    grad_accumulation_steps: 1
+    run_dir: work_dirs
+    type: VLAMetric
+    window_size: 1
+  mixed_precision_dtype: bf16
+  sampler: null
+  sharding_strategy: full-shard
+  type: FSDPTrainRunner
+  warmup_ratio: 0.05
+  weight_decay: 1.0e-05
+train_dataloader:
+  dataset:
+    datasets:
+      action_key: action
+      action_window_size: 20
+      data_root_path: ./datasets/libero_goal_no_noops_lerobotv2.1
+      frame_window_size: 17
+      statistic_name: libero_goal_no_noops
+      transforms:
+      - embodiment_id: 0
+        name_mappings:
+          actions:
+          - actions
+          observation.state:
+          - states
+        parquet_keys:
+        - observation.state
+        - timestamp
+        - actions
+        - info
+        - stats
+        - action_masks
+        type: ProcessParquetInputs
+        video_keys:
+        - observation.images.image
+        - observation.images.wrist_image
+      - type: ParquetPrompter
+        use_conversation: false
+      - max_len: 512
+        tokenizer:
+          model_path: ./checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl
+          type: PretrainedTokenizer
+        type: ProcessPrompts
+      - height: 128
+        type: ResizeImages
+        width: 128
+      - type: SimpleNormalizeImages
+      - action_dim: 32
+        action_key: action
+        norm_type: mean_std
+        state_dim: 32
+        state_key: proprio
+        type: NormalizeStatesAndActions
+      - frame_window_size: 17
+        num_views: 2
+        type: PrepareVideo
+      type: ParquetDataset
+      use_delta: false
+      window_start_idx: 0
+    name_mappings:
+      action:
+      - action
+      observation.state:
+      - proprio
+    statistic_keys:
+    - observation.state
+    - timestamp
+    - action
+    statistic_name: libero_goal_no_noops
+    type: DistributedRepeatingDataset
+  per_device_batch_size: 4
+  per_device_num_workers: 4

dreamzero_libero_goal_full_finetune_w_cache_bs64/dataset_statistics.json

@@ -0,0 +1,138 @@
+{
+  "libero_goal_no_noops": {
+    "proprio": {
+      "mean": [
+        -0.09891432758878499,
+        0.01489584178884142,
+        1.067322519907531,
+        2.8289916028867363,
+        0.31907902813946676,
+        -0.2782741362135122,
+        0.02821884616217808,
+        -0.02719759491844234
+      ],
+      "std": [
+        0.04140466877133655,
+        0.04117920944028847,
+        0.037538661473221754,
+        0.1955033740750787,
+        0.2574273555156476,
+        0.12620857483432396,
+        0.0053229405182870755,
+        0.005316267734440966
+      ],
+      "min": [
+        -0.46141287684440613,
+        -0.30136311054229736,
+        0.9083037972450256,
+        1.002794623374939,
+        -1.0517308712005615,
+        -1.5227035284042358,
+        -0.0021671096328645945,
+        -0.042015016078948975
+      ],
+      "max": [
+        0.13241106271743774,
+        0.3271525800228119,
+        1.472778081893921,
+        3.4731650352478027,
+        2.676265239715576,
+        0.6698114275932312,
+        0.04232141748070717,
+        0.001021005678921938
+      ],
+      "q01": [
+        -0.22800911694063627,
+        -0.10299188974829282,
+        0.9455820491176684,
+        2.608259821258135,
+        -0.18011099436472794,
+        -0.5618953405895196,
+        0.012391739034799726,
+        -0.039969403267763146
+      ],
+      "q99": [
+        -0.01061282617817482,
+        0.11139527808191847,
+        1.2117906450921032,
+        3.168615021869246,
+        0.6706572281431679,
+        0.05441701961452796,
+        0.04007683324960615,
+        -0.009863127064877238
+      ]
+    },
+    "timestamp": {
+      "mean": [
+        3.354542962472823
+      ],
+      "std": [
+        2.391036718656464
+      ],
+      "min": [
+        0.0
+      ],
+      "max": [
+        17.3
+      ],
+      "q01": null,
+      "q99": null
+    },
+    "action": {
+      "mean": [
+        0.04244028081958392,
+        0.03443110282231447,
+        -0.15229553502677498,
+        -0.0024877518145540465,
+        0.02584054000286765,
+        0.026984970605938637,
+        0.6345212227794035
+      ],
+      "std": [
+        0.15081003273695404,
+        0.13262089326077886,
+        0.18549323777289492,
+        0.020653428159559374,
+        0.029405301079120767,
+        0.03768659701327122,
+        0.1820141409830254
+      ],
+      "min": [
+        -0.9375,
+        -0.9375,
+        -0.9375,
+        -0.24214285612106323,
+        -0.375,
+        -0.2871428430080414,
+        0.0
+      ],
+      "max": [
+        0.9375,
+        0.9375,
+        0.9375,
+        0.3557142913341522,
+        0.375,
+        0.375,
+        1.0
+      ],
+      "q01": [
+        -0.6000398242585724,
+        -0.4840628973395442,
+        -0.828284557604454,
+        -0.08435403729704126,
+        -0.10102247173430137,
+        -0.08468755117369905,
+        0.22234615748180359
+      ],
+      "q99": [
+        0.7428644945783142,
+        0.5763548859545187,
+        0.6755035821278529,
+        0.09240677347557615,
+        0.17272708537297377,
+        0.14638082286503087,
+        1.0
+      ]
+    }
+  }
+}

dreamzero_libero_goal_full_finetune_w_cache_bs64/run-metrics.jsonl

@@ -0,0 +1 @@
+{"hparams": "{'_ckpt_root': './checkpoints', '_tokenizer': './checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl', '_frame_window_size': 17, 'model': {'type': 'DreamZeroVLA', 'num_views': 2, 'frame_window_size': 17, 'pretrained_name_or_path': './checkpoints/DreamZero-AgiBot', 'use_cache': True, 'vlm_backbone': {'type': 'WanBackbone', 'text_encoder_path': None, 'image_encoder_path': None, 'vae_path': None, 'tiled': False}, 'vla_head': {'type': 'DreamZeroHead', 'action_dim': 7, 'max_action_dim': 32, 'action_horizon': 10, 'max_state_dim': 64, 'num_frames': 17, 'num_frame_per_block': 2, 'num_action_per_block': 10, 'num_state_per_block': 1, 'frame_seqlen': 128, 'hidden_size': 1024, 'input_embedding_dim': 1536, 'dit_dim': 5120, 'dit_ffn_dim': 13824, 'dit_num_heads': 40, 'dit_num_layers': 40, 'dit_freq_dim': 256, 'dit_in_dim': 36, 'dit_out_dim': 16, 'max_num_embodiments': 32, 'noise_beta_alpha': 1.5, 'noise_beta_beta': 1.0, 'noise_s': 0.999, 'decouple_video_action_noise': False, 'video_noise_beta_alpha': 3.0, 'video_noise_beta_beta': 1.0, 'decouple_inference_noise': False, 'video_inference_final_noise': 0.8, 'num_inference_steps': 16, 'use_gradient_checkpointing': True, 'cfg_scale': 5.0, 'max_chunk_size': 2}, 'name_mapping': {'vla_head.model': 'action_head.model', 'vlm_backbone.text_encoder': 'action_head.text_encoder', 'vlm_backbone.image_encoder': 'action_head.image_encoder', 'vlm_backbone.vae': 'action_head.vae'}}, 'train_dataloader': {'per_device_batch_size': 4, 'per_device_num_workers': 4, 'dataset': {'type': 'DistributedRepeatingDataset', 'name_mappings': {'observation.state': ['proprio'], 'action': ['action']}, 'statistic_keys': ['observation.state', 'timestamp', 'action'], 'statistic_name': 'libero_goal_no_noops', 'datasets': {'type': 'ParquetDataset', 'data_root_path': './datasets/libero_goal_no_noops_lerobotv2.1', 'transforms': [{'type': 'ProcessParquetInputs', 'parquet_keys': ['observation.state', 'timestamp', 'actions', 'info', 'stats', 'action_masks'], 'video_keys': ['observation.images.image', 'observation.images.wrist_image'], 'name_mappings': {'observation.state': ['states'], 'actions': ['actions']}, 'embodiment_id': 0}, {'type': 'ParquetPrompter', 'use_conversation': False}, {'type': 'ProcessPrompts', 'tokenizer': {'type': 'PretrainedTokenizer', 'model_path': './checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl'}, 'max_len': 512}, {'type': 'ResizeImages', 'height': 128, 'width': 128}, {'type': 'SimpleNormalizeImages'}, {'type': 'NormalizeStatesAndActions', 'action_dim': 32, 'state_dim': 32, 'state_key': 'proprio', 'action_key': 'action', 'norm_type': 'mean_std'}, {'type': 'PrepareVideo', 'num_views': 2, 'frame_window_size': 17}], 'action_window_size': 20, 'action_key': 'action', 'use_delta': False, 'statistic_name': 'libero_goal_no_noops', 'window_start_idx': 0, 'frame_window_size': 17}}}, 'runner': {'type': 'FSDPTrainRunner', 'max_epochs': 10, 'learning_rate': 1e-05, 'weight_decay': 1e-05, 'max_grad_norm': 1.0, 'collator': {'type': 'DictCollator', 'keys': ['states', 'images', 'img_masks', 'actions', 'action_masks', 'embodiment_ids', 'frame_masks', 'lang_tokens', 'lang_masks'], 'meta_keys': ['task_description', 'prompt', 'info', 'stats', 'timestamp']}, 'sampler': None, 'metric': {'type': 'VLAMetric', 'active_trackers': ('jsonl', 'wandb'), 'run_dir': 'work_dirs/dreamzero_libero_goal_full_finetune_w_cache_31ae5d6d_bs64', 'grad_accumulation_steps': 1, 'window_size': 1, 'hparams': Config (path: configs/dreamzero/dreamzero_libero_goal_full_finetune_w_cache.py): {'_ckpt_root': './checkpoints', '_tokenizer': './checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl', '_frame_window_size': 17, 'model': {'type': 'DreamZeroVLA', 'num_views': 2, 'frame_window_size': 17, 'pretrained_name_or_path': './checkpoints/DreamZero-AgiBot', 'use_cache': True, 'vlm_backbone': {'type': 'WanBackbone', 'text_encoder_path': None, 'image_encoder_path': None, 'vae_path': None, 'tiled': False}, 'vla_head': {'type': 'DreamZeroHead', 'action_dim': 7, 'max_action_dim': 32, 'action_horizon': 10, 'max_state_dim': 64, 'num_frames': 17, 'num_frame_per_block': 2, 'num_action_per_block': 10, 'num_state_per_block': 1, 'frame_seqlen': 128, 'hidden_size': 1024, 'input_embedding_dim': 1536, 'dit_dim': 5120, 'dit_ffn_dim': 13824, 'dit_num_heads': 40, 'dit_num_layers': 40, 'dit_freq_dim': 256, 'dit_in_dim': 36, 'dit_out_dim': 16, 'max_num_embodiments': 32, 'noise_beta_alpha': 1.5, 'noise_beta_beta': 1.0, 'noise_s': 0.999, 'decouple_video_action_noise': False, 'video_noise_beta_alpha': 3.0, 'video_noise_beta_beta': 1.0, 'decouple_inference_noise': False, 'video_inference_final_noise': 0.8, 'num_inference_steps': 16, 'use_gradient_checkpointing': True, 'cfg_scale': 5.0, 'max_chunk_size': 2}, 'name_mapping': {'vla_head.model': 'action_head.model', 'vlm_backbone.text_encoder': 'action_head.text_encoder', 'vlm_backbone.image_encoder': 'action_head.image_encoder', 'vlm_backbone.vae': 'action_head.vae'}}, 'train_dataloader': {'per_device_batch_size': 4, 'per_device_num_workers': 4, 'dataset': {'type': 'DistributedRepeatingDataset', 'name_mappings': {'observation.state': ['proprio'], 'action': ['action']}, 'statistic_keys': ['observation.state', 'timestamp', 'action'], 'statistic_name': 'libero_goal_no_noops', 'datasets': {'type': 'ParquetDataset', 'data_root_path': './datasets/libero_goal_no_noops_lerobotv2.1', 'transforms': [{'type': 'ProcessParquetInputs', 'parquet_keys': ['observation.state', 'timestamp', 'actions', 'info', 'stats', 'action_masks'], 'video_keys': ['observation.images.image', 'observation.images.wrist_image'], 'name_mappings': {'observation.state': ['states'], 'actions': ['actions']}, 'embodiment_id': 0}, {'type': 'ParquetPrompter', 'use_conversation': False}, {'type': 'ProcessPrompts', 'tokenizer': {'type': 'PretrainedTokenizer', 'model_path': './checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl'}, 'max_len': 512}, {'type': 'ResizeImages', 'height': 128, 'width': 128}, {'type': 'SimpleNormalizeImages'}, {'type': 'NormalizeStatesAndActions', 'action_dim': 32, 'state_dim': 32, 'state_key': 'proprio', 'action_key': 'action', 'norm_type': 'mean_std'}, {'type': 'PrepareVideo', 'num_views': 2, 'frame_window_size': 17}], 'action_window_size': 20, 'action_key': 'action', 'use_delta': False, 'statistic_name': 'libero_goal_no_noops', 'window_start_idx': 0, 'frame_window_size': 17}}}, 'runner': {'type': 'FSDPTrainRunner', 'max_epochs': 10, 'learning_rate': 1e-05, 'weight_decay': 1e-05, 'max_grad_norm': 1.0, 'collator': {'type': 'DictCollator', 'keys': ['states', 'images', 'img_masks', 'actions', 'action_masks', 'embodiment_ids', 'frame_masks', 'lang_tokens', 'lang_masks'], 'meta_keys': ['task_description', 'prompt', 'info', 'stats', 'timestamp']}, 'sampler': None, 'metric': {'type': 'VLAMetric', 'active_trackers': ('jsonl', 'wandb'), 'run_dir': 'work_dirs/dreamzero_libero_goal_full_finetune_w_cache_31ae5d6d_bs64', 'grad_accumulation_steps': 1, 'window_size': 1, 'hparams': Config (path: configs/dreamzero/dreamzero_libero_goal_full_finetune_w_cache.py): {...}, 'run_id': 'dreamzero_libero_goal_full_finetune_w_cache_2026_05_20_17_04_59'}, 'lr_scheduler_type': 'linear-warmup+cosine-decay', 'warmup_ratio': 0.05, 'enable_gradient_checkpointing': True, 'enable_mixed_precision_training': True, 'mixed_precision_dtype': 'bf16', 'sharding_strategy': 'full-shard', 'change_key_name': False, 'max_keep_ckpts': 1, 'cfg': Config (path: configs/dreamzero/dreamzero_libero_goal_full_finetune_w_cache.py): {...}, 'args': Namespace(config='configs/dreamzero/dreamzero_libero_goal_full_finetune_w_cache.py', work_dir='work_dirs/dreamzero_libero_goal_full_finetune_w_cache_31ae5d6d_bs64', cfg_options={'train_dataloader.per_device_batch_size': 4, 'per_device_num_workers': 4, 'runner.max_keep_ckpts': 1}, eval_after_train=True, resume_from=None)}, 'eval': {'type': 'LiberoEvalRunner', 'task_suite_name': 'libero_goal', 'model_family': 'dreamzero', 'eval_chunk_size': 10, 'resize_size': 128, 'num_trials_per_task': 50, 'num_steps_wait': 10, 'seed': 7, 'enable_mixed_precision_training': True, 'mixed_precision_dtype': 'bf16', 'dataset': {'type': 'LiberoParquetEvalDataset', 'img_buffer_len': 4, 'transforms': [{'type': 'ProcessLiberoEvalInputs', 'img_keys': ['agentview_image', 'robot0_eye_in_hand_image']}, {'type': 'TransformImage', 'image_resize_strategy': 'resize-naive', 'input_sizes': [[3, 128, 128], [3, 128, 128]], 'means': [[127.5, 127.5, 127.5], [127.5, 127.5, 127.5]], 'stds': [[127.5, 127.5, 127.5], [127.5, 127.5, 127.5]]}, {'type': 'LiberoProprioFromInputs', 'norm_type': 'mean_std', 'pos_key': 'robot0_eef_pos', 'quat_key': 'robot0_eef_quat', 'gripper_key': 'robot0_gripper_qpos', 'state_dim': 32, 'out_key': 'states'}, {'type': 'LiberoPromptFromInputs', 'tokenizer': {'type': 'PretrainedTokenizer', 'model_path': './checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl'}, 'max_len': 512, 'negative_prompt': 'Vibrant colors, overexposed, static, blurry details, text, subtitles, style, artwork, painting, image, still, grayscale, dull, worst quality, low quality, JPEG artifacts, ugly, mutilated, extra fingers, bad hands, bad face, deformed, disfigured, mutated limbs, fused fingers, stagnant image, cluttered background, three legs, many people in the background, walking backwards.', 'use_conversation': False}, {'type': 'PrepareVideo', 'num_views': 2, 'frame_window_size': 1}]}, 'denormalize_action': {'type': 'DenormalizeLiberoAction', 'norm_type': 'mean_std', 'action_dim': 7}}, 'per_device_num_workers': 4}, 'run_id': 'dreamzero_libero_goal_full_finetune_w_cache_2026_05_20_17_04_59'}, 'lr_scheduler_type': 'linear-warmup+cosine-decay', 'warmup_ratio': 0.05, 'enable_gradient_checkpointing': True, 'enable_mixed_precision_training': True, 'mixed_precision_dtype': 'bf16', 'sharding_strategy': 'full-shard', 'change_key_name': False, 'max_keep_ckpts': 1, 'cfg': Config (path: configs/dreamzero/dreamzero_libero_goal_full_finetune_w_cache.py): {'_ckpt_root': './checkpoints', '_tokenizer': './checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl', '_frame_window_size': 17, 'model': {'type': 'DreamZeroVLA', 'num_views': 2, 'frame_window_size': 17, 'pretrained_name_or_path': './checkpoints/DreamZero-AgiBot', 'use_cache': True, 'vlm_backbone': {'type': 'WanBackbone', 'text_encoder_path': None, 'image_encoder_path': None, 'vae_path': None, 'tiled': False}, 'vla_head': {'type': 'DreamZeroHead', 'action_dim': 7, 'max_action_dim': 32, 'action_horizon': 10, 'max_state_dim': 64, 'num_frames': 17, 'num_frame_per_block': 2, 'num_action_per_block': 10, 'num_state_per_block': 1, 'frame_seqlen': 128, 'hidden_size': 1024, 'input_embedding_dim': 1536, 'dit_dim': 5120, 'dit_ffn_dim': 13824, 'dit_num_heads': 40, 'dit_num_layers': 40, 'dit_freq_dim': 256, 'dit_in_dim': 36, 'dit_out_dim': 16, 'max_num_embodiments': 32, 'noise_beta_alpha': 1.5, 'noise_beta_beta': 1.0, 'noise_s': 0.999, 'decouple_video_action_noise': False, 'video_noise_beta_alpha': 3.0, 'video_noise_beta_beta': 1.0, 'decouple_inference_noise': False, 'video_inference_final_noise': 0.8, 'num_inference_steps': 16, 'use_gradient_checkpointing': True, 'cfg_scale': 5.0, 'max_chunk_size': 2}, 'name_mapping': {'vla_head.model': 'action_head.model', 'vlm_backbone.text_encoder': 'action_head.text_encoder', 'vlm_backbone.image_encoder': 'action_head.image_encoder', 'vlm_backbone.vae': 'action_head.vae'}}, 'train_dataloader': {'per_device_batch_size': 4, 'per_device_num_workers': 4, 'dataset': {'type': 'DistributedRepeatingDataset', 'name_mappings': {'observation.state': ['proprio'], 'action': ['action']}, 'statistic_keys': ['observation.state', 'timestamp', 'action'], 'statistic_name': 'libero_goal_no_noops', 'datasets': {'type': 'ParquetDataset', 'data_root_path': './datasets/libero_goal_no_noops_lerobotv2.1', 'transforms': [{'type': 'ProcessParquetInputs', 'parquet_keys': ['observation.state', 'timestamp', 'actions', 'info', 'stats', 'action_masks'], 'video_keys': ['observation.images.image', 'observation.images.wrist_image'], 'name_mappings': {'observation.state': ['states'], 'actions': ['actions']}, 'embodiment_id': 0}, {'type': 'ParquetPrompter', 'use_conversation': False}, {'type': 'ProcessPrompts', 'tokenizer': {'type': 'PretrainedTokenizer', 'model_path': './checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl'}, 'max_len': 512}, {'type': 'ResizeImages', 'height': 128, 'width': 128}, {'type': 'SimpleNormalizeImages'}, {'type': 'NormalizeStatesAndActions', 'action_dim': 32, 'state_dim': 32, 'state_key': 'proprio', 'action_key': 'action', 'norm_type': 'mean_std'}, {'type': 'PrepareVideo', 'num_views': 2, 'frame_window_size': 17}], 'action_window_size': 20, 'action_key': 'action', 'use_delta': False, 'statistic_name': 'libero_goal_no_noops', 'window_start_idx': 0, 'frame_window_size': 17}}}, 'runner': {'type': 'FSDPTrainRunner', 'max_epochs': 10, 'learning_rate': 1e-05, 'weight_decay': 1e-05, 'max_grad_norm': 1.0, 'collator': {'type': 'DictCollator', 'keys': ['states', 'images', 'img_masks', 'actions', 'action_masks', 'embodiment_ids', 'frame_masks', 'lang_tokens', 'lang_masks'], 'meta_keys': ['task_description', 'prompt', 'info', 'stats', 'timestamp']}, 'sampler': None, 'metric': {'type': 'VLAMetric', 'active_trackers': ('jsonl', 'wandb'), 'run_dir': 'work_dirs/dreamzero_libero_goal_full_finetune_w_cache_31ae5d6d_bs64', 'grad_accumulation_steps': 1, 'window_size': 1, 'hparams': Config (path: configs/dreamzero/dreamzero_libero_goal_full_finetune_w_cache.py): {...}, 'run_id': 'dreamzero_libero_goal_full_finetune_w_cache_2026_05_20_17_04_59'}, 'lr_scheduler_type': 'linear-warmup+cosine-decay', 'warmup_ratio': 0.05, 'enable_gradient_checkpointing': True, 'enable_mixed_precision_training': True, 'mixed_precision_dtype': 'bf16', 'sharding_strategy': 'full-shard', 'change_key_name': False, 'max_keep_ckpts': 1, 'cfg': Config (path: configs/dreamzero/dreamzero_libero_goal_full_finetune_w_cache.py): {...}, 'args': Namespace(config='configs/dreamzero/dreamzero_libero_goal_full_finetune_w_cache.py', work_dir='work_dirs/dreamzero_libero_goal_full_finetune_w_cache_31ae5d6d_bs64', cfg_options={'train_dataloader.per_device_batch_size': 4, 'per_device_num_workers': 4, 'runner.max_keep_ckpts': 1}, eval_after_train=True, resume_from=None)}, 'eval': {'type': 'LiberoEvalRunner', 'task_suite_name': 'libero_goal', 'model_family': 'dreamzero', 'eval_chunk_size': 10, 'resize_size': 128, 'num_trials_per_task': 50, 'num_steps_wait': 10, 'seed': 7, 'enable_mixed_precision_training': True, 'mixed_precision_dtype': 'bf16', 'dataset': {'type': 'LiberoParquetEvalDataset', 'img_buffer_len': 4, 'transforms': [{'type': 'ProcessLiberoEvalInputs', 'img_keys': ['agentview_image', 'robot0_eye_in_hand_image']}, {'type': 'TransformImage', 'image_resize_strategy': 'resize-naive', 'input_sizes': [[3, 128, 128], [3, 128, 128]], 'means': [[127.5, 127.5, 127.5], [127.5, 127.5, 127.5]], 'stds': [[127.5, 127.5, 127.5], [127.5, 127.5, 127.5]]}, {'type': 'LiberoProprioFromInputs', 'norm_type': 'mean_std', 'pos_key': 'robot0_eef_pos', 'quat_key': 'robot0_eef_quat', 'gripper_key': 'robot0_gripper_qpos', 'state_dim': 32, 'out_key': 'states'}, {'type': 'LiberoPromptFromInputs', 'tokenizer': {'type': 'PretrainedTokenizer', 'model_path': './checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl'}, 'max_len': 512, 'negative_prompt': 'Vibrant colors, overexposed, static, blurry details, text, subtitles, style, artwork, painting, image, still, grayscale, dull, worst quality, low quality, JPEG artifacts, ugly, mutilated, extra fingers, bad hands, bad face, deformed, disfigured, mutated limbs, fused fingers, stagnant image, cluttered background, three legs, many people in the background, walking backwards.', 'use_conversation': False}, {'type': 'PrepareVideo', 'num_views': 2, 'frame_window_size': 1}]}, 'denormalize_action': {'type': 'DenormalizeLiberoAction', 'norm_type': 'mean_std', 'action_dim': 7}}, 'per_device_num_workers': 4}, 'args': Namespace(config='configs/dreamzero/dreamzero_libero_goal_full_finetune_w_cache.py', work_dir='work_dirs/dreamzero_libero_goal_full_finetune_w_cache_31ae5d6d_bs64', cfg_options={'train_dataloader.per_device_batch_size': 4, 'per_device_num_workers': 4, 'runner.max_keep_ckpts': 1}, eval_after_train=True, resume_from=None)}, 'eval': {'type': 'LiberoEvalRunner', 'task_suite_name': 'libero_goal', 'model_family': 'dreamzero', 'eval_chunk_size': 10, 'resize_size': 128, 'num_trials_per_task': 50, 'num_steps_wait': 10, 'seed': 7, 'enable_mixed_precision_training': True, 'mixed_precision_dtype': 'bf16', 'dataset': {'type': 'LiberoParquetEvalDataset', 'img_buffer_len': 4, 'transforms': [{'type': 'ProcessLiberoEvalInputs', 'img_keys': ['agentview_image', 'robot0_eye_in_hand_image']}, {'type': 'TransformImage', 'image_resize_strategy': 'resize-naive', 'input_sizes': [[3, 128, 128], [3, 128, 128]], 'means': [[127.5, 127.5, 127.5], [127.5, 127.5, 127.5]], 'stds': [[127.5, 127.5, 127.5], [127.5, 127.5, 127.5]]}, {'type': 'LiberoProprioFromInputs', 'norm_type': 'mean_std', 'pos_key': 'robot0_eef_pos', 'quat_key': 'robot0_eef_quat', 'gripper_key': 'robot0_gripper_qpos', 'state_dim': 32, 'out_key': 'states'}, {'type': 'LiberoPromptFromInputs', 'tokenizer': {'type': 'PretrainedTokenizer', 'model_path': './checkpoints/Wan2.1-I2V-14B-480P/google/umt5-xxl'}, 'max_len': 512, 'negative_prompt': 'Vibrant colors, overexposed, static, blurry details, text, subtitles, style, artwork, painting, image, still, grayscale, dull, worst quality, low quality, JPEG artifacts, ugly, mutilated, extra fingers, bad hands, bad face, deformed, disfigured, mutated limbs, fused fingers, stagnant image, cluttered background, three legs, many people in the background, walking backwards.', 'use_conversation': False}, {'type': 'PrepareVideo', 'num_views': 2, 'frame_window_size': 1}]}, 'denormalize_action': {'type': 'DenormalizeLiberoAction', 'norm_type': 'mean_std', 'action_dim': 7}}, 'per_device_num_workers': 4}", "run_id": "dreamzero_libero_goal_full_finetune_w_cache_2026_05_20_17_04_59"}