Upload folder using huggingface_hub

openvla_dino_siglip_llama2_libero_spatial_full_finetune_bs64/checkpoints/step-019872-epoch-24-loss=0.3600.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47fa787741a766f882899b4308d3b27c9eba8ad121cc3a76fe430c0fe2e4d91b
+size 30165078344

openvla_dino_siglip_llama2_libero_spatial_full_finetune_bs64/config.json

@@ -0,0 +1,407 @@
+{
+  "eval": {
+    "dataset": {
+      "transforms": [
+        {
+          "img_keys": [
+            "agentview_image",
+            "agentview_image"
+          ],
+          "type": "ProcessLiberoEvalInputs"
+        },
+        {
+          "image_resize_strategy": "resize-naive",
+          "input_sizes": [
+            [
+              3,
+              224,
+              224
+            ],
+            [
+              3,
+              224,
+              224
+            ]
+          ],
+          "means": [
+            [
+              123.515625,
+              116.04492188,
+              103.59375
+            ],
+            [
+              128,
+              128,
+              128
+            ]
+          ],
+          "stds": [
+            [
+              58.27148438,
+              57.02636719,
+              57.27539062
+            ],
+            [
+              128,
+              128,
+              128
+            ]
+          ],
+          "type": "TransformImage"
+        },
+        {
+          "max_len": null,
+          "prompt_suffix": " ",
+          "tokenizer": {
+            "model_path": "./checkpoints/openvla-7b-finetuned-libero-10",
+            "type": "PretrainedTokenizer"
+          },
+          "type": "LiberoPromptFromInputs"
+        }
+      ],
+      "type": "LiberoParquetEvalDataset"
+    },
+    "denormalize_action": {
+      "action_norm_mask": [
+        true,
+        true,
+        true,
+        true,
+        true,
+        true,
+        false
+      ],
+      "norm_type": "quantile",
+      "type": "DenormalizeLiberoAction"
+    },
+    "model_family": "openvla",
+    "num_steps_wait": 10,
+    "num_trials_per_task": 50,
+    "resize_size": 224,
+    "seed": 7,
+    "task_suite_name": "libero_spatial",
+    "type": "LiberoEvalRunner"
+  },
+  "model": {
+    "arch_specifier": "no-align+fused-gelu-mlp",
+    "freeze_llm_backbone": false,
+    "freeze_projector": false,
+    "freeze_vision_backbone": false,
+    "llm_backbone": {
+      "hf_token": null,
+      "inference_mode": false,
+      "llm_backbone_id": "llama2-7b-pure_causal",
+      "llm_family": "llama",
+      "llm_max_length": 2048,
+      "llm_path": "./checkpoints/Llama-2-7b-hf",
+      "type": "LLaMa2LLMBackbone"
+    },
+    "name_mapping": {
+      "llm_backbone.llm": "language_model",
+      "vision_backbone.dino_featurizer": "vision_backbone.featurizer",
+      "vision_backbone.siglip_featurizer": "vision_backbone.fused_featurizer"
+    },
+    "pretrained_name_or_path": "./checkpoints/openvla-7b-finetuned-libero-10",
+    "projector": {
+      "fused_vision_dim": 2176,
+      "llm_dim": 4096,
+      "type": "FusedMLPProjector"
+    },
+    "tokenizer": {
+      "bins": 256,
+      "max_action": 1,
+      "min_action": -1,
+      "model_path": "checkpoints/openvla-7b-finetuned-libero-10",
+      "type": "ActionTokenizer"
+    },
+    "type": "OpenVLA",
+    "vision_backbone": {
+      "dino_config": {
+        "file": "./checkpoints/vit_large_patch14_reg4_dinov2.lvd142m/model.safetensors",
+        "model_id": "dino"
+      },
+      "image_resize_strategy": "resize-naive",
+      "siglip_config": {
+        "file": "./checkpoints/ViT-SO400M-14-SigLIP/open_clip_model.safetensors",
+        "model_id": "siglip_224"
+      },
+      "type": "DinoSigLIPViTBackbone",
+      "vision_backbone_id": "dinosiglip-vit-so-224px"
+    },
+    "vla_head": {
+      "norm_stats": null,
+      "type": "OpenVLAHead",
+      "vocab_size": 32000
+    }
+  },
+  "per_device_num_workers": 8,
+  "runner": {
+    "collator": {
+      "ignore_idx": -100,
+      "model_max_length": 2048,
+      "pad_token_id": 0,
+      "padding_side": "right",
+      "pixel_values_dtype": "fp16",
+      "type": "PaddedCollatorForActionPrediction"
+    },
+    "enable_gradient_checkpointing": false,
+    "enable_mixed_precision_training": true,
+    "eval": {
+      "dataset": {
+        "transforms": [
+          {
+            "img_keys": [
+              "agentview_image",
+              "agentview_image"
+            ],
+            "type": "ProcessLiberoEvalInputs"
+          },
+          {
+            "image_resize_strategy": "resize-naive",
+            "input_sizes": [
+              [
+                3,
+                224,
+                224
+              ],
+              [
+                3,
+                224,
+                224
+              ]
+            ],
+            "means": [
+              [
+                123.515625,
+                116.04492188,
+                103.59375
+              ],
+              [
+                128,
+                128,
+                128
+              ]
+            ],
+            "stds": [
+              [
+                58.27148438,
+                57.02636719,
+                57.27539062
+              ],
+              [
+                128,
+                128,
+                128
+              ]
+            ],
+            "type": "TransformImage"
+          },
+          {
+            "max_len": null,
+            "prompt_suffix": " ",
+            "tokenizer": {
+              "model_path": "openvla/openvla-7b-finetuned-libero-10",
+              "type": "PretrainedTokenizer"
+            },
+            "type": "LiberoPromptFromInputs"
+          }
+        ],
+        "type": "LiberoParquetEvalDataset"
+      },
+      "denormalize_action": {
+        "action_norm_mask": [
+          true,
+          true,
+          true,
+          true,
+          true,
+          true,
+          false
+        ],
+        "norm_type": "quantile",
+        "type": "DenormalizeLiberoAction"
+      },
+      "model_family": "openvla",
+      "num_steps_wait": 10,
+      "num_trials_per_task": 50,
+      "resize_size": 224,
+      "seed": 7,
+      "task_suite_name": "libero_spatial",
+      "type": "LiberoEvalRunner"
+    },
+    "learning_rate": 2e-05,
+    "lr_scheduler_type": "constant",
+    "max_epochs": 24,
+    "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,
+    "type": "FSDPTrainRunner",
+    "warmup_ratio": 0.0,
+    "weight_decay": 0.0
+  },
+  "train_dataloader": {
+    "dataset": {
+      "datasets": {
+        "action_key": "action",
+        "action_window_size": 1,
+        "data_root_path": "./datasets/libero_spatial_no_noops_lerobotv2.1",
+        "repeat_to_full_length": true,
+        "statistic_name": "libero_spatial_no_noops",
+        "train_episode_fraction": 0.95,
+        "transforms": [
+          {
+            "dataset_name": "libero_spatial_no_noops",
+            "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.image"
+            ]
+          },
+          {
+            "action_dim": 7,
+            "action_key": "action",
+            "action_norm_mask": [
+              true,
+              true,
+              true,
+              true,
+              true,
+              true,
+              false
+            ],
+            "action_norm_type": "quantile",
+            "clip_norm": true,
+            "norm_type": "quantile",
+            "state_key": "proprio",
+            "state_norm_type": "min_max",
+            "type": "NormalizeStatesAndActions"
+          },
+          {
+            "action_tokenizer": {
+              "bins": 256,
+              "max_action": 1,
+              "min_action": -1,
+              "model_path": "./checkpoints/openvla-7b-finetuned-libero-10",
+              "type": "ActionTokenizer"
+            },
+            "lowercase_task_description": true,
+            "type": "ParquetPrompter"
+          },
+          {
+            "max_len": null,
+            "tokenizer": {
+              "model_path": "./checkpoints/openvla-7b-finetuned-libero-10",
+              "type": "PretrainedTokenizer"
+            },
+            "type": "ProcessPrompts",
+            "with_labels": true
+          },
+          {
+            "height": 224,
+            "type": "ResizeImagesLanczos",
+            "width": 224
+          },
+          {
+            "means": [
+              [
+                123.515625,
+                116.04492188,
+                103.59375
+              ],
+              [
+                128,
+                128,
+                128
+              ]
+            ],
+            "stds": [
+              [
+                58.27148438,
+                57.02636719,
+                57.27539062
+              ],
+              [
+                128,
+                128,
+                128
+              ]
+            ],
+            "type": "NormalizeImages"
+          }
+        ],
+        "type": "ParquetDataset",
+        "use_delta": false,
+        "window_start_idx": 0
+      },
+      "name_mappings": {
+        "action": [
+          "action"
+        ],
+        "observation.state": [
+          "proprio"
+        ]
+      },
+      "reshuffle_each_epoch": true,
+      "statistic_keys": [
+        "observation.state",
+        "timestamp",
+        "action"
+      ],
+      "statistic_name": "libero_spatial_no_noops",
+      "statistics_overrides": {
+        "libero_spatial_no_noops": {
+          "action": {
+            "q01": [
+              -0.7454732114076613,
+              -0.6616071462631226,
+              -0.9375,
+              -0.1071428582072258,
+              -0.20678570866584778,
+              -0.1842857152223587,
+              0.0
+            ],
+            "q99": [
+              0.9375,
+              0.8758928775787354,
+              0.9321428537368774,
+              0.1039285734295845,
+              0.17678570747375488,
+              0.14571428298950195,
+              1.0
+            ]
+          }
+        }
+      },
+      "type": "DistributedRepeatingDataset"
+    },
+    "per_device_batch_size": 8,
+    "per_device_num_workers": 4
+  }
+}

openvla_dino_siglip_llama2_libero_spatial_full_finetune_bs64/config.yaml

@@ -0,0 +1,293 @@
+eval:
+  dataset:
+    transforms:
+    - img_keys:
+      - agentview_image
+      - agentview_image
+      type: ProcessLiberoEvalInputs
+    - image_resize_strategy: resize-naive
+      input_sizes:
+      - - 3
+        - 224
+        - 224
+      - - 3
+        - 224
+        - 224
+      means:
+      - - 123.515625
+        - 116.04492188
+        - 103.59375
+      - - 128
+        - 128
+        - 128
+      stds:
+      - - 58.27148438
+        - 57.02636719
+        - 57.27539062
+      - - 128
+        - 128
+        - 128
+      type: TransformImage
+    - max_len: null
+      prompt_suffix: ' '
+      tokenizer:
+        model_path: ./checkpoints/openvla-7b-finetuned-libero-10
+        type: PretrainedTokenizer
+      type: LiberoPromptFromInputs
+    type: LiberoParquetEvalDataset
+  denormalize_action:
+    action_norm_mask:
+    - true
+    - true
+    - true
+    - true
+    - true
+    - true
+    - false
+    norm_type: quantile
+    type: DenormalizeLiberoAction
+  model_family: openvla
+  num_steps_wait: 10
+  num_trials_per_task: 50
+  resize_size: 224
+  seed: 7
+  task_suite_name: libero_spatial
+  type: LiberoEvalRunner
+model:
+  arch_specifier: no-align+fused-gelu-mlp
+  freeze_llm_backbone: false
+  freeze_projector: false
+  freeze_vision_backbone: false
+  llm_backbone:
+    hf_token: null
+    inference_mode: false
+    llm_backbone_id: llama2-7b-pure_causal
+    llm_family: llama
+    llm_max_length: 2048
+    llm_path: ./checkpoints/Llama-2-7b-hf
+    type: LLaMa2LLMBackbone
+  name_mapping:
+    llm_backbone.llm: language_model
+    vision_backbone.dino_featurizer: vision_backbone.featurizer
+    vision_backbone.siglip_featurizer: vision_backbone.fused_featurizer
+  pretrained_name_or_path: ./checkpoints/openvla-7b-finetuned-libero-10
+  projector:
+    fused_vision_dim: 2176
+    llm_dim: 4096
+    type: FusedMLPProjector
+  tokenizer:
+    bins: 256
+    max_action: 1
+    min_action: -1
+    model_path: checkpoints/openvla-7b-finetuned-libero-10
+    type: ActionTokenizer
+  type: OpenVLA
+  vision_backbone:
+    dino_config:
+      file: ./checkpoints/vit_large_patch14_reg4_dinov2.lvd142m/model.safetensors
+      model_id: dino
+    image_resize_strategy: resize-naive
+    siglip_config:
+      file: ./checkpoints/ViT-SO400M-14-SigLIP/open_clip_model.safetensors
+      model_id: siglip_224
+    type: DinoSigLIPViTBackbone
+    vision_backbone_id: dinosiglip-vit-so-224px
+  vla_head:
+    norm_stats: null
+    type: OpenVLAHead
+    vocab_size: 32000
+per_device_num_workers: 8
+runner:
+  collator:
+    ignore_idx: -100
+    model_max_length: 2048
+    pad_token_id: 0
+    padding_side: right
+    pixel_values_dtype: fp16
+    type: PaddedCollatorForActionPrediction
+  enable_gradient_checkpointing: false
+  enable_mixed_precision_training: true
+  eval:
+    dataset:
+      transforms:
+      - img_keys:
+        - agentview_image
+        - agentview_image
+        type: ProcessLiberoEvalInputs
+      - image_resize_strategy: resize-naive
+        input_sizes:
+        - - 3
+          - 224
+          - 224
+        - - 3
+          - 224
+          - 224
+        means:
+        - - 123.515625
+          - 116.04492188
+          - 103.59375
+        - - 128
+          - 128
+          - 128
+        stds:
+        - - 58.27148438
+          - 57.02636719
+          - 57.27539062
+        - - 128
+          - 128
+          - 128
+        type: TransformImage
+      - max_len: null
+        prompt_suffix: ' '
+        tokenizer:
+          model_path: openvla/openvla-7b-finetuned-libero-10
+          type: PretrainedTokenizer
+        type: LiberoPromptFromInputs
+      type: LiberoParquetEvalDataset
+    denormalize_action:
+      action_norm_mask:
+      - true
+      - true
+      - true
+      - true
+      - true
+      - true
+      - false
+      norm_type: quantile
+      type: DenormalizeLiberoAction
+    model_family: openvla
+    num_steps_wait: 10
+    num_trials_per_task: 50
+    resize_size: 224
+    seed: 7
+    task_suite_name: libero_spatial
+    type: LiberoEvalRunner
+  learning_rate: 2.0e-05
+  lr_scheduler_type: constant
+  max_epochs: 24
+  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
+  type: FSDPTrainRunner
+  warmup_ratio: 0.0
+  weight_decay: 0.0
+train_dataloader:
+  dataset:
+    datasets:
+      action_key: action
+      action_window_size: 1
+      data_root_path: ./datasets/libero_spatial_no_noops_lerobotv2.1
+      repeat_to_full_length: true
+      statistic_name: libero_spatial_no_noops
+      train_episode_fraction: 0.95
+      transforms:
+      - dataset_name: libero_spatial_no_noops
+        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.image
+      - action_dim: 7
+        action_key: action
+        action_norm_mask:
+        - true
+        - true
+        - true
+        - true
+        - true
+        - true
+        - false
+        action_norm_type: quantile
+        clip_norm: true
+        norm_type: quantile
+        state_key: proprio
+        state_norm_type: min_max
+        type: NormalizeStatesAndActions
+      - action_tokenizer:
+          bins: 256
+          max_action: 1
+          min_action: -1
+          model_path: ./checkpoints/openvla-7b-finetuned-libero-10
+          type: ActionTokenizer
+        lowercase_task_description: true
+        type: ParquetPrompter
+      - max_len: null
+        tokenizer:
+          model_path: ./checkpoints/openvla-7b-finetuned-libero-10
+          type: PretrainedTokenizer
+        type: ProcessPrompts
+        with_labels: true
+      - height: 224
+        type: ResizeImagesLanczos
+        width: 224
+      - means:
+        - - 123.515625
+          - 116.04492188
+          - 103.59375
+        - - 128
+          - 128
+          - 128
+        stds:
+        - - 58.27148438
+          - 57.02636719
+          - 57.27539062
+        - - 128
+          - 128
+          - 128
+        type: NormalizeImages
+      type: ParquetDataset
+      use_delta: false
+      window_start_idx: 0
+    name_mappings:
+      action:
+      - action
+      observation.state:
+      - proprio
+    reshuffle_each_epoch: true
+    statistic_keys:
+    - observation.state
+    - timestamp
+    - action
+    statistic_name: libero_spatial_no_noops
+    statistics_overrides:
+      libero_spatial_no_noops:
+        action:
+          q01:
+          - -0.7454732114076613
+          - -0.6616071462631226
+          - -0.9375
+          - -0.1071428582072258
+          - -0.20678570866584778
+          - -0.1842857152223587
+          - 0.0
+          q99:
+          - 0.9375
+          - 0.8758928775787354
+          - 0.9321428537368774
+          - 0.1039285734295845
+          - 0.17678570747375488
+          - 0.14571428298950195
+          - 1.0
+    type: DistributedRepeatingDataset
+  per_device_batch_size: 8
+  per_device_num_workers: 4

openvla_dino_siglip_llama2_libero_spatial_full_finetune_bs64/dataset_statistics.json

@@ -0,0 +1,120 @@
+{
+  "libero_spatial_no_noops": {
+    "proprio": {
+      "mean": [
+        -0.024462566693947342,
+        0.10653030478388664,
+        1.0580495716300307,
+        3.062855007870368,
+        -0.10464045916348884,
+        0.08307320236969534,
+        0.019954609054627596,
+        -0.02016269208612657
+      ],
+      "std": [
+        0.03894316411915835,
+        0.04873628070932603,
+        0.03692094784082842,
+        0.0369502396792774,
+        0.14538513627309543,
+        0.07695788947742314,
+        0.006102641532497049,
+        0.006049884044502419
+      ],
+      "min": [
+        -0.3095473051071167,
+        -0.29250794649124146,
+        0.9095591306686401,
+        2.497488260269165,
+        -1.8006486892700195,
+        -0.7207611203193665,
+        -0.0004703797458205372,
+        -0.041536275297403336
+      ],
+      "max": [
+        0.1759040206670761,
+        0.3904820382595062,
+        1.3290715217590332,
+        3.4566118717193604,
+        1.2268599271774292,
+        1.0429412126541138,
+        0.041053611785173416,
+        0.000775813648942858
+      ],
+      "q01": null,
+      "q99": null
+    },
+    "timestamp": {
+      "mean": [
+        3.1281914291108173
+      ],
+      "std": [
+        1.9190018719668336
+      ],
+      "min": [
+        0.0
+      ],
+      "max": [
+        9.6
+      ],
+      "q01": null,
+      "q99": null
+    },
+    "action": {
+      "mean": [
+        0.15312488430795423,
+        0.13707241597825376,
+        -0.15526779033841448,
+        -0.005176474488725037,
+        -0.011208756940533639,
+        -0.02019425420384803,
+        0.4578818200364616
+      ],
+      "std": [
+        0.15599777645651164,
+        0.13125442554438385,
+        0.19226097301543327,
+        0.014084604392168992,
+        0.02738322007326005,
+        0.021779582921450876,
+        0.18831055119433956
+      ],
+      "min": [
+        -0.9375,
+        -0.9375,
+        -0.9375,
+        -0.1875,
+        -0.3675000071525574,
+        -0.36000001430511475,
+        0.0
+      ],
+      "max": [
+        0.9375,
+        0.9375,
+        0.9375,
+        0.1971428543329239,
+        0.33642858266830444,
+        0.375,
+        1.0
+      ],
+      "q01": [
+        -0.7454732114076613,
+        -0.6616071462631226,
+        -0.9375,
+        -0.1071428582072258,
+        -0.20678570866584778,
+        -0.1842857152223587,
+        0.0
+      ],
+      "q99": [
+        0.9375,
+        0.8758928775787354,
+        0.9321428537368774,
+        0.1039285734295845,
+        0.17678570747375488,
+        0.14571428298950195,
+        1.0
+      ]
+    }
+  }
+}

openvla_dino_siglip_llama2_libero_spatial_full_finetune_bs64/llm_backbone_config.json

@@ -0,0 +1,30 @@
+{
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "pad_token_id": 0,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float32",
+  "transformers_version": "4.53.2",
+  "use_cache": false,
+  "vocab_size": 32064
+}

openvla_dino_siglip_llama2_libero_spatial_full_finetune_bs64/run-metrics.jsonl

@@ -0,0 +1 @@
+{"hparams": "{'model': {'arch_specifier': 'no-align+fused-gelu-mlp', 'freeze_llm_backbone': False, 'freeze_projector': False, 'freeze_vision_backbone': False, 'llm_backbone': {'hf_token': None, 'inference_mode': False, 'llm_backbone_id': 'llama2-7b-pure_causal', 'llm_family': 'llama', 'llm_max_length': 2048, 'llm_path': './checkpoints/Llama-2-7b-hf', 'type': 'LLaMa2LLMBackbone'}, 'name_mapping': {'llm_backbone.llm': 'language_model', 'vision_backbone.dino_featurizer': 'vision_backbone.featurizer', 'vision_backbone.siglip_featurizer': 'vision_backbone.fused_featurizer'}, 'pretrained_name_or_path': './checkpoints/openvla-7b-finetuned-libero-10', 'projector': {'fused_vision_dim': 2176, 'llm_dim': 4096, 'type': 'FusedMLPProjector'}, 'tokenizer': {'bins': 256, 'max_action': 1, 'min_action': -1, 'model_path': 'checkpoints/openvla-7b-finetuned-libero-10', 'type': 'ActionTokenizer'}, 'type': 'OpenVLA', 'vision_backbone': {'dino_config': {'file': './checkpoints/vit_large_patch14_reg4_dinov2.lvd142m/model.safetensors', 'model_id': 'dino'}, 'image_resize_strategy': 'resize-naive', 'siglip_config': {'file': './checkpoints/ViT-SO400M-14-SigLIP/open_clip_model.safetensors', 'model_id': 'siglip_224'}, 'type': 'DinoSigLIPViTBackbone', 'vision_backbone_id': 'dinosiglip-vit-so-224px'}, 'vla_head': {'norm_stats': None, 'type': 'OpenVLAHead', 'vocab_size': 32000}}, 'train_dataloader': {'dataset': {'datasets': {'action_key': 'action', 'action_window_size': 1, 'data_root_path': './datasets/libero_spatial_no_noops_lerobotv2.1', 'repeat_to_full_length': True, 'statistic_name': 'libero_spatial_no_noops', 'train_episode_fraction': 0.95, 'transforms': [{'dataset_name': 'libero_spatial_no_noops', '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.image']}, {'action_dim': 7, 'action_key': 'action', 'action_norm_mask': [True, True, True, True, True, True, False], 'action_norm_type': 'quantile', 'clip_norm': True, 'norm_type': 'quantile', 'state_key': 'proprio', 'state_norm_type': 'min_max', 'type': 'NormalizeStatesAndActions'}, {'action_tokenizer': {'bins': 256, 'max_action': 1, 'min_action': -1, 'model_path': './checkpoints/openvla-7b-finetuned-libero-10', 'type': 'ActionTokenizer'}, 'lowercase_task_description': True, 'type': 'ParquetPrompter'}, {'max_len': None, 'tokenizer': {'model_path': './checkpoints/openvla-7b-finetuned-libero-10', 'type': 'PretrainedTokenizer'}, 'type': 'ProcessPrompts', 'with_labels': True}, {'height': 224, 'type': 'ResizeImagesLanczos', 'width': 224}, {'means': [[123.515625, 116.04492188, 103.59375], [128, 128, 128]], 'stds': [[58.27148438, 57.02636719, 57.27539062], [128, 128, 128]], 'type': 'NormalizeImages'}], 'type': 'ParquetDataset', 'use_delta': False, 'window_start_idx': 0}, 'name_mappings': {'action': ['action'], 'observation.state': ['proprio']}, 'reshuffle_each_epoch': True, 'statistics_overrides': {'libero_spatial_no_noops': {'action': {'q01': [-0.7454732114076613, -0.6616071462631226, -0.9375, -0.1071428582072258, -0.20678570866584778, -0.1842857152223587, 0.0], 'q99': [0.9375, 0.8758928775787354, 0.9321428537368774, 0.1039285734295845, 0.17678570747375488, 0.14571428298950195, 1.0]}}}, 'statistic_keys': ['observation.state', 'timestamp', 'action'], 'statistic_name': 'libero_spatial_no_noops', 'type': 'DistributedRepeatingDataset'}, 'per_device_batch_size': 8, 'per_device_num_workers': 4}, 'runner': {'collator': {'ignore_idx': -100, 'model_max_length': 2048, 'pad_token_id': 0, 'padding_side': 'right', 'pixel_values_dtype': 'fp16', 'type': 'PaddedCollatorForActionPrediction'}, 'enable_gradient_checkpointing': False, 'enable_mixed_precision_training': True, 'eval': {'dataset': {'transforms': [{'img_keys': ['agentview_image', 'agentview_image'], 'type': 'ProcessLiberoEvalInputs'}, {'image_resize_strategy': 'resize-naive', 'input_sizes': [[3, 224, 224], [3, 224, 224]], 'means': [[123.515625, 116.04492188, 103.59375], [128, 128, 128]], 'stds': [[58.27148438, 57.02636719, 57.27539062], [128, 128, 128]], 'type': 'TransformImage'}, {'max_len': None, 'prompt_suffix': ' ', 'tokenizer': {'model_path': 'openvla/openvla-7b-finetuned-libero-10', 'type': 'PretrainedTokenizer'}, 'type': 'LiberoPromptFromInputs'}], 'type': 'LiberoParquetEvalDataset'}, 'denormalize_action': {'action_norm_mask': [True, True, True, True, True, True, False], 'norm_type': 'quantile', 'type': 'DenormalizeLiberoAction'}, 'model_family': 'openvla', 'num_steps_wait': 10, 'num_trials_per_task': 50, 'resize_size': 224, 'seed': 7, 'task_suite_name': 'libero_spatial', 'type': 'LiberoEvalRunner'}, 'learning_rate': 2e-05, 'lr_scheduler_type': 'constant', 'max_epochs': 24, 'max_grad_norm': 1.0, 'metric': {'active_trackers': ('jsonl', 'wandb'), 'grad_accumulation_steps': 1, 'run_dir': 'work_dirs/openvla_dino_siglip_llama2_libero_spatial_full_finetune_ba9d13a4_bs64', 'type': 'VLAMetric', 'window_size': 1, 'hparams': Config (path: configs/openvla/openvla_dino_siglip_llama2_libero_spatial_full_finetune.py): {'model': {'arch_specifier': 'no-align+fused-gelu-mlp', 'freeze_llm_backbone': False, 'freeze_projector': False, 'freeze_vision_backbone': False, 'llm_backbone': {'hf_token': None, 'inference_mode': False, 'llm_backbone_id': 'llama2-7b-pure_causal', 'llm_family': 'llama', 'llm_max_length': 2048, 'llm_path': './checkpoints/Llama-2-7b-hf', 'type': 'LLaMa2LLMBackbone'}, 'name_mapping': {'llm_backbone.llm': 'language_model', 'vision_backbone.dino_featurizer': 'vision_backbone.featurizer', 'vision_backbone.siglip_featurizer': 'vision_backbone.fused_featurizer'}, 'pretrained_name_or_path': './checkpoints/openvla-7b-finetuned-libero-10', 'projector': {'fused_vision_dim': 2176, 'llm_dim': 4096, 'type': 'FusedMLPProjector'}, 'tokenizer': {'bins': 256, 'max_action': 1, 'min_action': -1, 'model_path': 'checkpoints/openvla-7b-finetuned-libero-10', 'type': 'ActionTokenizer'}, 'type': 'OpenVLA', 'vision_backbone': {'dino_config': {'file': './checkpoints/vit_large_patch14_reg4_dinov2.lvd142m/model.safetensors', 'model_id': 'dino'}, 'image_resize_strategy': 'resize-naive', 'siglip_config': {'file': './checkpoints/ViT-SO400M-14-SigLIP/open_clip_model.safetensors', 'model_id': 'siglip_224'}, 'type': 'DinoSigLIPViTBackbone', 'vision_backbone_id': 'dinosiglip-vit-so-224px'}, 'vla_head': {'norm_stats': None, 'type': 'OpenVLAHead', 'vocab_size': 32000}}, 'train_dataloader': {'dataset': {'datasets': {'action_key': 'action', 'action_window_size': 1, 'data_root_path': './datasets/libero_spatial_no_noops_lerobotv2.1', 'repeat_to_full_length': True, 'statistic_name': 'libero_spatial_no_noops', 'train_episode_fraction': 0.95, 'transforms': [{'dataset_name': 'libero_spatial_no_noops', '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.image']}, {'action_dim': 7, 'action_key': 'action', 'action_norm_mask': [True, True, True, True, True, True, False], 'action_norm_type': 'quantile', 'clip_norm': True, 'norm_type': 'quantile', 'state_key': 'proprio', 'state_norm_type': 'min_max', 'type': 'NormalizeStatesAndActions'}, {'action_tokenizer': {'bins': 256, 'max_action': 1, 'min_action': -1, 'model_path': './checkpoints/openvla-7b-finetuned-libero-10', 'type': 'ActionTokenizer'}, 'lowercase_task_description': True, 'type': 'ParquetPrompter'}, {'max_len': None, 'tokenizer': {'model_path': './checkpoints/openvla-7b-finetuned-libero-10', 'type': 'PretrainedTokenizer'}, 'type': 'ProcessPrompts', 'with_labels': True}, {'height': 224, 'type': 'ResizeImagesLanczos', 'width': 224}, {'means': [[123.515625, 116.04492188, 103.59375], [128, 128, 128]], 'stds': [[58.27148438, 57.02636719, 57.27539062], [128, 128, 128]], 'type': 'NormalizeImages'}], 'type': 'ParquetDataset', 'use_delta': False, 'window_start_idx': 0}, 'name_mappings': {'action': ['action'], 'observation.state': ['proprio']}, 'reshuffle_each_epoch': True, 'statistics_overrides': {'libero_spatial_no_noops': {'action': {'q01': [-0.7454732114076613, -0.6616071462631226, -0.9375, -0.1071428582072258, -0.20678570866584778, -0.1842857152223587, 0.0], 'q99': [0.9375, 0.8758928775787354, 0.9321428537368774, 0.1039285734295845, 0.17678570747375488, 0.14571428298950195, 1.0]}}}, 'statistic_keys': ['observation.state', 'timestamp', 'action'], 'statistic_name': 'libero_spatial_no_noops', 'type': 'DistributedRepeatingDataset'}, 'per_device_batch_size': 8, 'per_device_num_workers': 4}, 'runner': {'collator': {'ignore_idx': -100, 'model_max_length': 2048, 'pad_token_id': 0, 'padding_side': 'right', 'pixel_values_dtype': 'fp16', 'type': 'PaddedCollatorForActionPrediction'}, 'enable_gradient_checkpointing': False, 'enable_mixed_precision_training': True, 'eval': {'dataset': {'transforms': [{'img_keys': ['agentview_image', 'agentview_image'], 'type': 'ProcessLiberoEvalInputs'}, {'image_resize_strategy': 'resize-naive', 'input_sizes': [[3, 224, 224], [3, 224, 224]], 'means': [[123.515625, 116.04492188, 103.59375], [128, 128, 128]], 'stds': [[58.27148438, 57.02636719, 57.27539062], [128, 128, 128]], 'type': 'TransformImage'}, {'max_len': None, 'prompt_suffix': ' ', 'tokenizer': {'model_path': 'openvla/openvla-7b-finetuned-libero-10', 'type': 'PretrainedTokenizer'}, 'type': 'LiberoPromptFromInputs'}], 'type': 'LiberoParquetEvalDataset'}, 'denormalize_action': {'action_norm_mask': [True, True, True, True, True, True, False], 'norm_type': 'quantile', 'type': 'DenormalizeLiberoAction'}, 'model_family': 'openvla', 'num_steps_wait': 10, 'num_trials_per_task': 50, 'resize_size': 224, 'seed': 7, 'task_suite_name': 'libero_spatial', 'type': 'LiberoEvalRunner'}, 'learning_rate': 2e-05, 'lr_scheduler_type': 'constant', 'max_epochs': 24, 'max_grad_norm': 1.0, 'metric': {'active_trackers': ('jsonl', 'wandb'), 'grad_accumulation_steps': 1, 'run_dir': 'work_dirs/openvla_dino_siglip_llama2_libero_spatial_full_finetune_ba9d13a4_bs64', 'type': 'VLAMetric', 'window_size': 1, 'hparams': Config (path: configs/openvla/openvla_dino_siglip_llama2_libero_spatial_full_finetune.py): {...}, 'run_id': 'openvla_dino_siglip_llama2_libero_spatial_full_finetune_2026_05_27_09_34_01'}, 'mixed_precision_dtype': 'bf16', 'sampler': None, 'type': 'FSDPTrainRunner', 'warmup_ratio': 0.0, 'weight_decay': 0.0, 'max_keep_ckpts': 1, 'cfg': Config (path: configs/openvla/openvla_dino_siglip_llama2_libero_spatial_full_finetune.py): {...}, 'args': Namespace(config='configs/openvla/openvla_dino_siglip_llama2_libero_spatial_full_finetune.py', work_dir='work_dirs/openvla_dino_siglip_llama2_libero_spatial_full_finetune_ba9d13a4_bs64', cfg_options={'train_dataloader.per_device_batch_size': 8, 'per_device_num_workers': 8, 'runner.max_keep_ckpts': 1}, eval_after_train=True, resume_from=None)}, 'eval': {'dataset': {'transforms': [{'img_keys': ['agentview_image', 'agentview_image'], 'type': 'ProcessLiberoEvalInputs'}, {'image_resize_strategy': 'resize-naive', 'input_sizes': [[3, 224, 224], [3, 224, 224]], 'means': [[123.515625, 116.04492188, 103.59375], [128, 128, 128]], 'stds': [[58.27148438, 57.02636719, 57.27539062], [128, 128, 128]], 'type': 'TransformImage'}, {'max_len': None, 'prompt_suffix': ' ', 'tokenizer': {'model_path': './checkpoints/openvla-7b-finetuned-libero-10', 'type': 'PretrainedTokenizer'}, 'type': 'LiberoPromptFromInputs'}], 'type': 'LiberoParquetEvalDataset'}, 'denormalize_action': {'action_norm_mask': [True, True, True, True, True, True, False], 'norm_type': 'quantile', 'type': 'DenormalizeLiberoAction'}, 'model_family': 'openvla', 'num_steps_wait': 10, 'num_trials_per_task': 50, 'resize_size': 224, 'seed': 7, 'task_suite_name': 'libero_spatial', 'type': 'LiberoEvalRunner'}, 'per_device_num_workers': 8}, 'run_id': 'openvla_dino_siglip_llama2_libero_spatial_full_finetune_2026_05_27_09_34_01'}, 'mixed_precision_dtype': 'bf16', 'sampler': None, 'type': 'FSDPTrainRunner', 'warmup_ratio': 0.0, 'weight_decay': 0.0, 'max_keep_ckpts': 1, 'cfg': Config (path: configs/openvla/openvla_dino_siglip_llama2_libero_spatial_full_finetune.py): {'model': {'arch_specifier': 'no-align+fused-gelu-mlp', 'freeze_llm_backbone': False, 'freeze_projector': False, 'freeze_vision_backbone': False, 'llm_backbone': {'hf_token': None, 'inference_mode': False, 'llm_backbone_id': 'llama2-7b-pure_causal', 'llm_family': 'llama', 'llm_max_length': 2048, 'llm_path': './checkpoints/Llama-2-7b-hf', 'type': 'LLaMa2LLMBackbone'}, 'name_mapping': {'llm_backbone.llm': 'language_model', 'vision_backbone.dino_featurizer': 'vision_backbone.featurizer', 'vision_backbone.siglip_featurizer': 'vision_backbone.fused_featurizer'}, 'pretrained_name_or_path': './checkpoints/openvla-7b-finetuned-libero-10', 'projector': {'fused_vision_dim': 2176, 'llm_dim': 4096, 'type': 'FusedMLPProjector'}, 'tokenizer': {'bins': 256, 'max_action': 1, 'min_action': -1, 'model_path': 'checkpoints/openvla-7b-finetuned-libero-10', 'type': 'ActionTokenizer'}, 'type': 'OpenVLA', 'vision_backbone': {'dino_config': {'file': './checkpoints/vit_large_patch14_reg4_dinov2.lvd142m/model.safetensors', 'model_id': 'dino'}, 'image_resize_strategy': 'resize-naive', 'siglip_config': {'file': './checkpoints/ViT-SO400M-14-SigLIP/open_clip_model.safetensors', 'model_id': 'siglip_224'}, 'type': 'DinoSigLIPViTBackbone', 'vision_backbone_id': 'dinosiglip-vit-so-224px'}, 'vla_head': {'norm_stats': None, 'type': 'OpenVLAHead', 'vocab_size': 32000}}, 'train_dataloader': {'dataset': {'datasets': {'action_key': 'action', 'action_window_size': 1, 'data_root_path': './datasets/libero_spatial_no_noops_lerobotv2.1', 'repeat_to_full_length': True, 'statistic_name': 'libero_spatial_no_noops', 'train_episode_fraction': 0.95, 'transforms': [{'dataset_name': 'libero_spatial_no_noops', '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.image']}, {'action_dim': 7, 'action_key': 'action', 'action_norm_mask': [True, True, True, True, True, True, False], 'action_norm_type': 'quantile', 'clip_norm': True, 'norm_type': 'quantile', 'state_key': 'proprio', 'state_norm_type': 'min_max', 'type': 'NormalizeStatesAndActions'}, {'action_tokenizer': {'bins': 256, 'max_action': 1, 'min_action': -1, 'model_path': './checkpoints/openvla-7b-finetuned-libero-10', 'type': 'ActionTokenizer'}, 'lowercase_task_description': True, 'type': 'ParquetPrompter'}, {'max_len': None, 'tokenizer': {'model_path': './checkpoints/openvla-7b-finetuned-libero-10', 'type': 'PretrainedTokenizer'}, 'type': 'ProcessPrompts', 'with_labels': True}, {'height': 224, 'type': 'ResizeImagesLanczos', 'width': 224}, {'means': [[123.515625, 116.04492188, 103.59375], [128, 128, 128]], 'stds': [[58.27148438, 57.02636719, 57.27539062], [128, 128, 128]], 'type': 'NormalizeImages'}], 'type': 'ParquetDataset', 'use_delta': False, 'window_start_idx': 0}, 'name_mappings': {'action': ['action'], 'observation.state': ['proprio']}, 'reshuffle_each_epoch': True, 'statistics_overrides': {'libero_spatial_no_noops': {'action': {'q01': [-0.7454732114076613, -0.6616071462631226, -0.9375, -0.1071428582072258, -0.20678570866584778, -0.1842857152223587, 0.0], 'q99': [0.9375, 0.8758928775787354, 0.9321428537368774, 0.1039285734295845, 0.17678570747375488, 0.14571428298950195, 1.0]}}}, 'statistic_keys': ['observation.state', 'timestamp', 'action'], 'statistic_name': 'libero_spatial_no_noops', 'type': 'DistributedRepeatingDataset'}, 'per_device_batch_size': 8, 'per_device_num_workers': 4}, 'runner': {'collator': {'ignore_idx': -100, 'model_max_length': 2048, 'pad_token_id': 0, 'padding_side': 'right', 'pixel_values_dtype': 'fp16', 'type': 'PaddedCollatorForActionPrediction'}, 'enable_gradient_checkpointing': False, 'enable_mixed_precision_training': True, 'eval': {'dataset': {'transforms': [{'img_keys': ['agentview_image', 'agentview_image'], 'type': 'ProcessLiberoEvalInputs'}, {'image_resize_strategy': 'resize-naive', 'input_sizes': [[3, 224, 224], [3, 224, 224]], 'means': [[123.515625, 116.04492188, 103.59375], [128, 128, 128]], 'stds': [[58.27148438, 57.02636719, 57.27539062], [128, 128, 128]], 'type': 'TransformImage'}, {'max_len': None, 'prompt_suffix': ' ', 'tokenizer': {'model_path': 'openvla/openvla-7b-finetuned-libero-10', 'type': 'PretrainedTokenizer'}, 'type': 'LiberoPromptFromInputs'}], 'type': 'LiberoParquetEvalDataset'}, 'denormalize_action': {'action_norm_mask': [True, True, True, True, True, True, False], 'norm_type': 'quantile', 'type': 'DenormalizeLiberoAction'}, 'model_family': 'openvla', 'num_steps_wait': 10, 'num_trials_per_task': 50, 'resize_size': 224, 'seed': 7, 'task_suite_name': 'libero_spatial', 'type': 'LiberoEvalRunner'}, 'learning_rate': 2e-05, 'lr_scheduler_type': 'constant', 'max_epochs': 24, 'max_grad_norm': 1.0, 'metric': {'active_trackers': ('jsonl', 'wandb'), 'grad_accumulation_steps': 1, 'run_dir': 'work_dirs/openvla_dino_siglip_llama2_libero_spatial_full_finetune_ba9d13a4_bs64', 'type': 'VLAMetric', 'window_size': 1, 'hparams': Config (path: configs/openvla/openvla_dino_siglip_llama2_libero_spatial_full_finetune.py): {...}, 'run_id': 'openvla_dino_siglip_llama2_libero_spatial_full_finetune_2026_05_27_09_34_01'}, 'mixed_precision_dtype': 'bf16', 'sampler': None, 'type': 'FSDPTrainRunner', 'warmup_ratio': 0.0, 'weight_decay': 0.0, 'max_keep_ckpts': 1, 'cfg': Config (path: configs/openvla/openvla_dino_siglip_llama2_libero_spatial_full_finetune.py): {...}, 'args': Namespace(config='configs/openvla/openvla_dino_siglip_llama2_libero_spatial_full_finetune.py', work_dir='work_dirs/openvla_dino_siglip_llama2_libero_spatial_full_finetune_ba9d13a4_bs64', cfg_options={'train_dataloader.per_device_batch_size': 8, 'per_device_num_workers': 8, 'runner.max_keep_ckpts': 1}, eval_after_train=True, resume_from=None)}, 'eval': {'dataset': {'transforms': [{'img_keys': ['agentview_image', 'agentview_image'], 'type': 'ProcessLiberoEvalInputs'}, {'image_resize_strategy': 'resize-naive', 'input_sizes': [[3, 224, 224], [3, 224, 224]], 'means': [[123.515625, 116.04492188, 103.59375], [128, 128, 128]], 'stds': [[58.27148438, 57.02636719, 57.27539062], [128, 128, 128]], 'type': 'TransformImage'}, {'max_len': None, 'prompt_suffix': ' ', 'tokenizer': {'model_path': './checkpoints/openvla-7b-finetuned-libero-10', 'type': 'PretrainedTokenizer'}, 'type': 'LiberoPromptFromInputs'}], 'type': 'LiberoParquetEvalDataset'}, 'denormalize_action': {'action_norm_mask': [True, True, True, True, True, True, False], 'norm_type': 'quantile', 'type': 'DenormalizeLiberoAction'}, 'model_family': 'openvla', 'num_steps_wait': 10, 'num_trials_per_task': 50, 'resize_size': 224, 'seed': 7, 'task_suite_name': 'libero_spatial', 'type': 'LiberoEvalRunner'}, 'per_device_num_workers': 8}, 'args': Namespace(config='configs/openvla/openvla_dino_siglip_llama2_libero_spatial_full_finetune.py', work_dir='work_dirs/openvla_dino_siglip_llama2_libero_spatial_full_finetune_ba9d13a4_bs64', cfg_options={'train_dataloader.per_device_batch_size': 8, 'per_device_num_workers': 8, 'runner.max_keep_ckpts': 1}, eval_after_train=True, resume_from=None)}, 'eval': {'dataset': {'transforms': [{'img_keys': ['agentview_image', 'agentview_image'], 'type': 'ProcessLiberoEvalInputs'}, {'image_resize_strategy': 'resize-naive', 'input_sizes': [[3, 224, 224], [3, 224, 224]], 'means': [[123.515625, 116.04492188, 103.59375], [128, 128, 128]], 'stds': [[58.27148438, 57.02636719, 57.27539062], [128, 128, 128]], 'type': 'TransformImage'}, {'max_len': None, 'prompt_suffix': ' ', 'tokenizer': {'model_path': './checkpoints/openvla-7b-finetuned-libero-10', 'type': 'PretrainedTokenizer'}, 'type': 'LiberoPromptFromInputs'}], 'type': 'LiberoParquetEvalDataset'}, 'denormalize_action': {'action_norm_mask': [True, True, True, True, True, True, False], 'norm_type': 'quantile', 'type': 'DenormalizeLiberoAction'}, 'model_family': 'openvla', 'num_steps_wait': 10, 'num_trials_per_task': 50, 'resize_size': 224, 'seed': 7, 'task_suite_name': 'libero_spatial', 'type': 'LiberoEvalRunner'}, 'per_device_num_workers': 8}", "run_id": "openvla_dino_siglip_llama2_libero_spatial_full_finetune_2026_05_27_09_34_01"}