Add model checkpoint

.gitattributes

@@ -0,0 +1 @@
+*.pt filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,30 @@
+---
+base_model:
+- openvla/openvla-7b-prismatic
+- CogACT/CogACT-Base
+tags:
+- robotics
+- vla
+- multimodal
+- pretraining
+pipeline_tag: robotics
+---
+# Model Card for CronusVLA
+
+**Weights**
+
+`step2/checkpoints/reshape_embedding_step2.pt`: 
+
+- (1) This checkpoint serves as the starting point for fine-tuning on the Libero benchmark. 
+- (2) It is a post-trained model derived from CronusVLA-7B, configured with a two-step frame (history length = 1).
+- (3) During the original post-training phase, the model was trained to predict future actions with a chunk size of 16. To ensure compatibility with the Libero setting of OpenVLA-oft (chunk = 8), we modified the position embeddings responsible for action chunking in the original checkpoint. 
+- (4) Note: This checkpoint is not directly evaluable and is intended only as a fine-tuning initialization.
+
+`step4/checkpoints/reshape_embedding_step4.pt`: 
+
+- (1) This checkpoint serves as the starting point for fine-tuning on the LIBERO benchmark. 
+- (2) It is a post-trained model derived from CronusVLA-7B, configured with a four-step frame (history length = 3).
+- (3) During the original post-training phase, the model was trained to predict future actions with a chunk size of 16. To ensure compatibility with the Libero setting of OpenVLA-oft (chunk = 8), we modified the position embeddings responsible for action chunking in the original checkpoint. 
+- (4) Note: This checkpoint is not directly evaluable and is intended only as a fine-tuning initialization.
+
+> If you want to use these checkpoint to finetune, please follow the instruction of [CronusVLA](https://github.com/InternRobotics/CronusVLA).

step2/README.md

@@ -0,0 +1,23 @@
+---
+base_model:
+- openvla/openvla-7b-prismatic
+- CogACT/CogACT-Base
+tags:
+- robotics
+- vla
+- multimodal
+- pretraining
+pipeline_tag: robotics
+---
+# Model Card for CronusVLA
+
+**Weights**
+
+`checkpoints/reshape_embedding_step2.pt`: 
+
+- (1) This checkpoint serves as the starting point for fine-tuning on the Libero benchmark. 
+- (2) It is a post-trained model derived from CronusVLA-7B, configured with a two-step frame (history length = 1).
+- (3) During the original post-training phase, the model was trained to predict future actions with a chunk size of 16. To ensure compatibility with the Libero setting of OpenVLA-oft (chunk = 8), we modified the position embeddings responsible for action chunking in the original checkpoint. 
+- (4) Note: This checkpoint is not directly evaluable and is intended only as a fine-tuning initialization.
+
+> If you want to use these checkpoint to finetune, please follow the instruction of [CronusVLA](https://github.com/InternRobotics/CronusVLA).

step2/checkpoints/reshape_embedding_step2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc133922d2b5fce6ef42476b4557d26de4789da444cf24afff1417dc05e9f612
+size 30703958828

step2/config.json

@@ -0,0 +1,52 @@
+{
+  "action_dim": 7,
+  "action_model_type": "DiT-B",
+  "data_root_dir": "",
+  "debug": false,
+  "future_action_window_size": 15,
+  "hf_token": "hf_token",
+  "image_aug": true,
+  "is_resume": false,
+  "load_all_data_for_training": true,
+  "past_action_window_size": 1,
+  "pretrained_checkpoint": "path/to/ckpt",
+  "repeated_diffusion_steps": 4,
+  "resume_epoch": null,
+  "resume_step": null,
+  "run_id": "step2_7B",
+  "run_id_note": null,
+  "run_root_dir": "outputs/step2_7B",
+  "save_interval": 2500,
+  "seed": 42,
+  "trackers": [
+    "jsonl",
+    "wandb"
+  ],
+  "use_ema": false,
+  "vla": {
+    "base_vlm": "prism-dinosiglip-224px+7b",
+    "data_mix": "bridge_rt_1",
+    "enable_gradient_checkpointing": true,
+    "enable_mixed_precision_training": true,
+    "epochs": 100,
+    "expected_world_size": 16,
+    "freeze_llm_backbone": false,
+    "freeze_vision_backbone": false,
+    "global_batch_size": 256,
+    "learning_rate": 2e-05,
+    "lr_scheduler_type": "constant",
+    "max_grad_norm": 1.0,
+    "max_steps": null,
+    "per_device_batch_size": 16,
+    "reduce_in_full_precision": true,
+    "shuffle_buffer_size": 250000,
+    "train_strategy": "fsdp-full-shard",
+    "type": "prism-dinosiglip-224px+oxe+diffusion",
+    "unfreeze_last_llm_layer": false,
+    "vla_id": "prism-dinosiglip-224px+oxe+diffusion",
+    "warmup_ratio": 0.0,
+    "weight_decay": 0.0
+  },
+  "wandb_entity": "",
+  "wandb_project": ""
+}

step2/config.yaml

@@ -0,0 +1,48 @@
+action_dim: 7
+action_model_type: DiT-B
+data_root_dir: ""
+debug: false
+future_action_window_size: 15
+hf_token: hf_token
+image_aug: true
+is_resume: false
+load_all_data_for_training: true
+past_action_window_size: 1
+pretrained_checkpoint: "path/to/ckpt"
+repeated_diffusion_steps: 4
+resume_epoch: null
+resume_step: null
+run_id: step2_7B
+run_id_note: null
+run_root_dir: outputs/step2_7B
+save_interval: 2500
+seed: 42
+trackers:
+- jsonl
+- wandb
+use_ema: false
+vla:
+  base_vlm: prism-dinosiglip-224px+7b
+  data_mix: bridge_rt_1
+  enable_gradient_checkpointing: true
+  enable_mixed_precision_training: true
+  epochs: 100
+  expected_world_size: 16
+  freeze_llm_backbone: false
+  freeze_vision_backbone: false
+  global_batch_size: 256
+  learning_rate: 2.0e-05
+  lr_scheduler_type: constant
+  max_grad_norm: 1.0
+  max_steps: null
+  per_device_batch_size: 16
+  reduce_in_full_precision: true
+  shuffle_buffer_size: 250000
+  train_strategy: fsdp-full-shard
+  type: prism-dinosiglip-224px+oxe+diffusion
+  unfreeze_last_llm_layer: false
+  vla_id: prism-dinosiglip-224px+oxe+diffusion
+  warmup_ratio: 0.0
+  weight_decay: 0.0
+wandb_entity: ""
+wandb_project: ""

step2/dataset_statistics.json

@@ -0,0 +1,252 @@
+{
+  "bridge_dataset": {
+    "action": {
+      "mean": [
+        0.00023341973428614438,
+        0.0001300475705647841,
+        -0.00012762400729116052,
+        -0.00015565499779768288,
+        -0.00040393517701886594,
+        0.0002355772303417325,
+        0.5764579772949219
+      ],
+      "std": [
+        0.009765934199094772,
+        0.013689189217984676,
+        0.012667394243180752,
+        0.028534121811389923,
+        0.030637938529253006,
+        0.07691467553377151,
+        0.4973696768283844
+      ],
+      "max": [
+        0.41691166162490845,
+        0.25864794850349426,
+        0.21218234300613403,
+        3.122201919555664,
+        1.8618112802505493,
+        6.280478477478027,
+        1.0
+      ],
+      "min": [
+        -0.4007510244846344,
+        -0.13874775171279907,
+        -0.22553899884223938,
+        -3.2010786533355713,
+        -1.8618112802505493,
+        -6.279075622558594,
+        0.0
+      ],
+      "q01": [
+        -0.02872725307941437,
+        -0.04170349963009357,
+        -0.026093858778476715,
+        -0.08092105075716972,
+        -0.09288699507713317,
+        -0.20718276381492615,
+        0.0
+      ],
+      "q99": [
+        0.028309678435325586,
+        0.040855254605412394,
+        0.040161586627364146,
+        0.08192047759890528,
+        0.07792850524187081,
+        0.20382574498653397,
+        1.0
+      ],
+      "mask": [
+        true,
+        true,
+        true,
+        true,
+        true,
+        true,
+        false
+      ]
+    },
+    "proprio": {
+      "mean": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "std": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "max": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "min": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q01": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q99": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ]
+    },
+    "num_transitions": 2135463,
+    "num_trajectories": 60064
+  },
+  "fractal20220817_data": {
+    "action": {
+      "mean": [
+        0.006987491622567177,
+        0.00626587588340044,
+        -0.012625089846551418,
+        0.04333178699016571,
+        -0.005756180267781019,
+        0.0009131028782576323,
+        0.5354204773902893
+      ],
+      "std": [
+        0.06921201944351196,
+        0.059655144810676575,
+        0.0735311210155487,
+        0.15610052645206451,
+        0.131641685962677,
+        0.14593306183815002,
+        0.49710750579833984
+      ],
+      "max": [
+        2.9984593391418457,
+        22.09052848815918,
+        2.7507524490356445,
+        1.570636510848999,
+        1.5321086645126343,
+        1.5691522359848022,
+        1.0
+      ],
+      "min": [
+        -2.0204520225524902,
+        -5.497899532318115,
+        -2.031663417816162,
+        -1.569917917251587,
+        -1.569892168045044,
+        -1.570419430732727,
+        0.0
+      ],
+      "q01": [
+        -0.22453527510166169,
+        -0.14820013284683228,
+        -0.231589707583189,
+        -0.3517994859814644,
+        -0.4193011274933815,
+        -0.43643461108207704,
+        0.0
+      ],
+      "q99": [
+        0.17824687153100965,
+        0.14938379630446405,
+        0.21842354819178575,
+        0.5892666035890578,
+        0.35272657424211445,
+        0.44796681255102094,
+        1.0
+      ],
+      "mask": [
+        true,
+        true,
+        true,
+        true,
+        true,
+        true,
+        false
+      ]
+    },
+    "proprio": {
+      "mean": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "std": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "max": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "min": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q01": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q99": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ]
+    },
+    "num_transitions": 3786400,
+    "num_trajectories": 87212
+  }
+}

step2/run-metrics.jsonl

@@ -0,0 +1 @@
+{"hparams": {"action_dim": 7, "action_model_type": "DiT-B", "data_root_dir": "", "debug": false, "future_action_window_size": 15, "hf_token": "hf_token", "image_aug": true, "is_resume": false, "load_all_data_for_training": true, "past_action_window_size": 1, "pretrained_checkpoint": "path/to/ckpt", "repeated_diffusion_steps": 4, "resume_epoch": null, "resume_step": null, "run_id": "step2_7B", "run_id_note": null, "run_root_dir": "outputs/step2_7B", "save_interval": 5000, "seed": 42, "trackers": ["jsonl", "wandb"], "use_ema": false, "vla": {"base_vlm": "prism-dinosiglip-224px+7b", "data_mix": "bridge_rt_1", "enable_gradient_checkpointing": true, "enable_mixed_precision_training": true, "epochs": 100, "expected_world_size": 16, "freeze_llm_backbone": false, "freeze_vision_backbone": false, "global_batch_size": 256, "learning_rate": 2e-05, "lr_scheduler_type": "constant", "max_grad_norm": 1.0, "max_steps": null, "per_device_batch_size": 16, "reduce_in_full_precision": true, "shuffle_buffer_size": 250000, "train_strategy": "fsdp-full-shard", "type": "prism-dinosiglip-224px+oxe+diffusion", "unfreeze_last_llm_layer": false, "vla_id": "prism-dinosiglip-224px+oxe+diffusion", "warmup_ratio": 0.0, "weight_decay": 0.0}, "wandb_entity": "", "wandb_project": ""}, "run_id": "step2_7B"}

step4/README.md

@@ -0,0 +1,23 @@
+---
+base_model:
+- openvla/openvla-7b-prismatic
+- CogACT/CogACT-Base
+tags:
+- robotics
+- vla
+- multimodal
+- pretraining
+pipeline_tag: robotics
+---
+# Model Card for CronusVLA
+
+**Weights**
+
+`step4/checkpoints/reshape_embedding_step4.pt`: 
+
+- (1) This checkpoint serves as the starting point for fine-tuning on the Libero benchmark. 
+- (2) It is a post-trained model derived from CronusVLA-7B, configured with a four-step frame (history length = 3).
+- (3) During the original post-training phase, the model was trained to predict future actions with a chunk size of 16. To ensure compatibility with the Libero setting of OpenVLA-oft (chunk = 8), we modified the position embeddings responsible for action chunking in the original checkpoint. 
+- (4) Note: This checkpoint is not directly evaluable and is intended only as a fine-tuning initialization.
+
+> If you want to use these checkpoint to finetune, please follow the instruction of [CronusVLA](https://github.com/InternRobotics/CronusVLA).

step4/checkpoints/reshape_embedding_step4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2142c243d7b87f29396bcedf242b2f5fd170bfca8895a773a70c5847c45a91c
+size 30703997740

step4/config.json

@@ -0,0 +1,52 @@
+{
+  "action_dim": 7,
+  "action_model_type": "DiT-B",
+  "data_root_dir": "",
+  "debug": false,
+  "future_action_window_size": 15,
+  "hf_token": "hf_token",
+  "image_aug": true,
+  "is_resume": false,
+  "load_all_data_for_training": true,
+  "past_action_window_size": 3,
+  "pretrained_checkpoint": "path/to/ckpt",
+  "repeated_diffusion_steps": 4,
+  "resume_epoch": null,
+  "resume_step": null,
+  "run_id": "step4_7B",
+  "run_id_note": null,
+  "run_root_dir": "outputs/step4_7B",
+  "save_interval": 2500,
+  "seed": 42,
+  "trackers": [
+    "jsonl",
+    "wandb"
+  ],
+  "use_ema": false,
+  "vla": {
+    "base_vlm": "prism-dinosiglip-224px+7b",
+    "data_mix": "bridge_rt_1",
+    "enable_gradient_checkpointing": true,
+    "enable_mixed_precision_training": true,
+    "epochs": 100,
+    "expected_world_size": 64,
+    "freeze_llm_backbone": false,
+    "freeze_vision_backbone": false,
+    "global_batch_size": 512,
+    "learning_rate": 4e-05,
+    "lr_scheduler_type": "constant",
+    "max_grad_norm": 1.0,
+    "max_steps": null,
+    "per_device_batch_size": 8,
+    "reduce_in_full_precision": true,
+    "shuffle_buffer_size": 250000,
+    "train_strategy": "fsdp-full-shard",
+    "type": "prism-dinosiglip-224px+oxe+diffusion",
+    "unfreeze_last_llm_layer": false,
+    "vla_id": "prism-dinosiglip-224px+oxe+diffusion",
+    "warmup_ratio": 0.0,
+    "weight_decay": 0.0
+  },
+  "wandb_entity": "",
+  "wandb_project": ""
+}

step4/config.yaml

@@ -0,0 +1,48 @@
+action_dim: 7
+action_model_type: DiT-B
+data_root_dir: ""
+debug: false
+future_action_window_size: 15
+hf_token: hf_token
+image_aug: true
+is_resume: false
+load_all_data_for_training: true
+past_action_window_size: 3
+pretrained_checkpoint: "path/to/ckpt"
+repeated_diffusion_steps: 4
+resume_epoch: null
+resume_step: null
+run_id: step4_7B
+run_id_note: null
+run_root_dir: outputs/step4_7B
+save_interval: 2500
+seed: 42
+trackers:
+- jsonl
+- wandb
+use_ema: false
+vla:
+  base_vlm: prism-dinosiglip-224px+7b
+  data_mix: bridge_rt_1
+  enable_gradient_checkpointing: true
+  enable_mixed_precision_training: true
+  epochs: 100
+  expected_world_size: 64
+  freeze_llm_backbone: false
+  freeze_vision_backbone: false
+  global_batch_size: 512
+  learning_rate: 4.0e-05
+  lr_scheduler_type: constant
+  max_grad_norm: 1.0
+  max_steps: null
+  per_device_batch_size: 8
+  reduce_in_full_precision: true
+  shuffle_buffer_size: 250000
+  train_strategy: fsdp-full-shard
+  type: prism-dinosiglip-224px+oxe+diffusion
+  unfreeze_last_llm_layer: false
+  vla_id: prism-dinosiglip-224px+oxe+diffusion
+  warmup_ratio: 0.0
+  weight_decay: 0.0
+wandb_entity: ""
+wandb_project: ""

step4/dataset_statistics.json

@@ -0,0 +1,252 @@
+{
+  "bridge_dataset": {
+    "action": {
+      "mean": [
+        0.00023341973428614438,
+        0.0001300475705647841,
+        -0.00012762400729116052,
+        -0.00015565499779768288,
+        -0.00040393517701886594,
+        0.0002355772303417325,
+        0.5764579772949219
+      ],
+      "std": [
+        0.009765934199094772,
+        0.013689189217984676,
+        0.012667394243180752,
+        0.028534121811389923,
+        0.030637938529253006,
+        0.07691467553377151,
+        0.4973696768283844
+      ],
+      "max": [
+        0.41691166162490845,
+        0.25864794850349426,
+        0.21218234300613403,
+        3.122201919555664,
+        1.8618112802505493,
+        6.280478477478027,
+        1.0
+      ],
+      "min": [
+        -0.4007510244846344,
+        -0.13874775171279907,
+        -0.22553899884223938,
+        -3.2010786533355713,
+        -1.8618112802505493,
+        -6.279075622558594,
+        0.0
+      ],
+      "q01": [
+        -0.02872725307941437,
+        -0.04170349963009357,
+        -0.026093858778476715,
+        -0.08092105075716972,
+        -0.09288699507713317,
+        -0.20718276381492615,
+        0.0
+      ],
+      "q99": [
+        0.028309678435325586,
+        0.040855254605412394,
+        0.040161586627364146,
+        0.08192047759890528,
+        0.07792850524187081,
+        0.20382574498653397,
+        1.0
+      ],
+      "mask": [
+        true,
+        true,
+        true,
+        true,
+        true,
+        true,
+        false
+      ]
+    },
+    "proprio": {
+      "mean": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "std": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "max": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "min": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q01": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q99": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ]
+    },
+    "num_transitions": 2135463,
+    "num_trajectories": 60064
+  },
+  "fractal20220817_data": {
+    "action": {
+      "mean": [
+        0.006987491622567177,
+        0.00626587588340044,
+        -0.012625089846551418,
+        0.04333178699016571,
+        -0.005756180267781019,
+        0.0009131028782576323,
+        0.5354204773902893
+      ],
+      "std": [
+        0.06921201944351196,
+        0.059655144810676575,
+        0.0735311210155487,
+        0.15610052645206451,
+        0.131641685962677,
+        0.14593306183815002,
+        0.49710750579833984
+      ],
+      "max": [
+        2.9984593391418457,
+        22.09052848815918,
+        2.7507524490356445,
+        1.570636510848999,
+        1.5321086645126343,
+        1.5691522359848022,
+        1.0
+      ],
+      "min": [
+        -2.0204520225524902,
+        -5.497899532318115,
+        -2.031663417816162,
+        -1.569917917251587,
+        -1.569892168045044,
+        -1.570419430732727,
+        0.0
+      ],
+      "q01": [
+        -0.22453527510166169,
+        -0.14820013284683228,
+        -0.231589707583189,
+        -0.3517994859814644,
+        -0.4193011274933815,
+        -0.43643461108207704,
+        0.0
+      ],
+      "q99": [
+        0.17824687153100965,
+        0.14938379630446405,
+        0.21842354819178575,
+        0.5892666035890578,
+        0.35272657424211445,
+        0.44796681255102094,
+        1.0
+      ],
+      "mask": [
+        true,
+        true,
+        true,
+        true,
+        true,
+        true,
+        false
+      ]
+    },
+    "proprio": {
+      "mean": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "std": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "max": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "min": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q01": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "q99": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ]
+    },
+    "num_transitions": 3786400,
+    "num_trajectories": 87212
+  }
+}

step4/run-metrics.jsonl

@@ -0,0 +1 @@
+{"hparams": {"action_dim": 7, "action_model_type": "DiT-B", "data_root_dir": "", "debug": false, "future_action_window_size": 15, "hf_token": "hf_token", "image_aug": true, "is_resume": false, "load_all_data_for_training": true, "past_action_window_size": 3, "pretrained_checkpoint": "path/to/ckpt", "repeated_diffusion_steps": 4, "resume_epoch": null, "resume_step": null, "run_id": "step4_7B", "run_id_note": null, "run_root_dir": "outputs/step4_7B", "save_interval": 2500, "seed": 42, "trackers": ["jsonl", "wandb"], "use_ema": false, "vla": {"base_vlm": "prism-dinosiglip-224px+7b", "data_mix": "bridge_rt_1", "enable_gradient_checkpointing": true, "enable_mixed_precision_training": true, "epochs": 100, "expected_world_size": 64, "freeze_llm_backbone": false, "freeze_vision_backbone": false, "global_batch_size": 512, "learning_rate": 4e-05, "lr_scheduler_type": "constant", "max_grad_norm": 1.0, "max_steps": null, "per_device_batch_size": 8, "reduce_in_full_precision": true, "shuffle_buffer_size": 250000, "train_strategy": "fsdp-full-shard", "type": "prism-dinosiglip-224px+oxe+diffusion", "unfreeze_last_llm_layer": false, "vla_id": "prism-dinosiglip-224px+oxe+diffusion", "warmup_ratio": 0.0, "weight_decay": 0.0}, "wandb_entity": "", "wandb_project": ""}, "run_id": "step4_7B"}