Add task-level fine-tuning-scope catalogue and preserve historical comparison limits

README.md

@@ -16,6 +16,7 @@ Inference checkpoints, original configurations and model provenance.
 | Visual history | [Visual history](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/visual-history) |
 | Latency in prompt | [Latency in prompt](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/latency-in-prompt) |
 | Latency transfer | [Latency transfer](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/latency-transfer) |
+| VLA fine-tuning scope | [VLA fine-tuning scope](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/vla-finetuning-scope) |
 | Mean vs. profile training | [Mean vs. profile training](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/mean-vs-profile) |
 
 | Resource | Repository |

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning/README.md

@@ -0,0 +1,5 @@
+# flappy_fix_latency_2_200ep_full_tuning
+
+[Fine-tuning settings and fixed asset references](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/vla-finetuning-scope/flappy)
+
+The original checkpoint, configuration and normalization statistics are retained. LoRA sources additionally require their recorded adapter and action-head files.

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning/artifact.json

@@ -0,0 +1,61 @@
+{
+  "run": "flappy_fix_latency_2_200ep_full_tuning",
+  "source": {
+    "repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+    "revision": "1ab96924e691cbe8924ac4671466723bdb5ec3af",
+    "path": "flappy_fix_latency_2_200ep_full_tuning"
+  },
+  "metadata": {
+    "repo": "latency-sensitive-bench/benchmark-models",
+    "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+    "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning"
+  },
+  "checkpoints": [
+    {
+      "repo": "latency-sensitive-bench/benchmark-models",
+      "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+      "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning/checkpoints/steps_5000_model.safetensors"
+    }
+  ],
+  "source_bindings": [
+    {
+      "run": "flappy_fix_latency_2_200ep_full_tuning",
+      "module": "vla-finetuning-scope",
+      "train_latency_raw_frames": 2,
+      "training_step": 5000,
+      "training_started_at_utc": "2026-06-23T16:41:09.236188Z",
+      "git_commit": "daeb076ebed09fb21a1a1b3ce6ab124c64018212",
+      "wandb_run_id": "hfsfeb3f",
+      "checkpoint": "flappy_fix_latency_2_200ep_full_tuning/checkpoints/steps_5000_model.safetensors",
+      "checkpoint_identity": "2ad30759d93084b8a8c6e03129e3239fac9a644b4c17c638deb76d82aa8610d5",
+      "source_path": "flappy_fix_latency_2_200ep_full_tuning/checkpoints/steps_5000_state/model.safetensors",
+      "source_model_revision": "946358fcf7850d259d33e1b822b80cf5aed15577",
+      "freeze_config": {
+        "freeze_modules": "",
+        "freeze_llm_bottom_ratio": 0.0
+      },
+      "dataset": {
+        "repo": "latency-sensitive-bench/benchmark-datasets",
+        "revision": "e0f4007f504978f93c32557ea4e9b97fdf62fcf5",
+        "path": "latency-aware/flappy/flappy_200ep/flappy_fix_latency_2_200ep",
+        "source_repo": "latency-sensitive-bench/flappy_200ep",
+        "source_revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+        "train_rows": 330734,
+        "train_episodes": 180,
+        "original_source": {
+          "repo": "latency-sensitive-bench/flappy_200ep",
+          "revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+          "path": "flappy_fix_latency_2_200ep"
+        }
+      },
+      "paper_consumers": [
+        "legacy/vla_hyper"
+      ]
+    }
+  ],
+  "evaluation": {
+    "repo": "latency-sensitive-bench/benchmark-datasets",
+    "revision": "58157c572dbd2e0e5036ce197c42db975e56ae4e",
+    "path": "evaluations/flappy/flappy_fix_latency_2_200ep_full_tuning"
+  }
+}

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected/README.md

@@ -0,0 +1,5 @@
+# flappy_fix_latency_2_200ep_full_tuning_corrected
+
+[Fine-tuning settings and fixed asset references](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/vla-finetuning-scope/flappy)
+
+The original checkpoint, configuration and normalization statistics are retained. LoRA sources additionally require their recorded adapter and action-head files.

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected/artifact.json

@@ -0,0 +1,61 @@
+{
+  "run": "flappy_fix_latency_2_200ep_full_tuning_corrected",
+  "source": {
+    "repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+    "revision": "1ab96924e691cbe8924ac4671466723bdb5ec3af",
+    "path": "flappy_fix_latency_2_200ep_full_tuning_corrected"
+  },
+  "metadata": {
+    "repo": "latency-sensitive-bench/benchmark-models",
+    "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+    "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected"
+  },
+  "checkpoints": [
+    {
+      "repo": "latency-sensitive-bench/benchmark-models",
+      "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+      "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected/checkpoints/steps_5000_model.safetensors"
+    }
+  ],
+  "source_bindings": [
+    {
+      "run": "flappy_fix_latency_2_200ep_full_tuning_corrected",
+      "module": "vla-finetuning-scope",
+      "train_latency_raw_frames": 2,
+      "training_step": 5000,
+      "training_started_at_utc": "2026-06-23T20:08:09.389236Z",
+      "git_commit": "b6f4f971098c0bc5e340ef616b823a4ca769c7e4",
+      "wandb_run_id": "jd1xm599",
+      "checkpoint": "flappy_fix_latency_2_200ep_full_tuning_corrected/checkpoints/steps_5000_model.safetensors",
+      "checkpoint_identity": "ff343b81e73667493a90c8c0696872ab7f3ed1bf55fae59cf8b57f9785b040eb",
+      "source_path": "flappy_fix_latency_2_200ep_full_tuning_corrected/checkpoints/steps_5000_state/model.safetensors",
+      "source_model_revision": "946358fcf7850d259d33e1b822b80cf5aed15577",
+      "freeze_config": {
+        "freeze_modules": "",
+        "freeze_llm_bottom_ratio": 0.0
+      },
+      "dataset": {
+        "repo": "latency-sensitive-bench/benchmark-datasets",
+        "revision": "e0f4007f504978f93c32557ea4e9b97fdf62fcf5",
+        "path": "latency-aware/flappy/flappy_200ep/flappy_fix_latency_2_200ep",
+        "source_repo": "latency-sensitive-bench/flappy_200ep",
+        "source_revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+        "train_rows": 330734,
+        "train_episodes": 180,
+        "original_source": {
+          "repo": "latency-sensitive-bench/flappy_200ep",
+          "revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+          "path": "flappy_fix_latency_2_200ep"
+        }
+      },
+      "paper_consumers": [
+        "exps/vla_finetuning_scope"
+      ]
+    }
+  ],
+  "evaluation": {
+    "repo": "latency-sensitive-bench/benchmark-datasets",
+    "revision": "58157c572dbd2e0e5036ce197c42db975e56ae4e",
+    "path": "evaluations/flappy/flappy_fix_latency_2_200ep_full_tuning_corrected"
+  }
+}

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps/README.md

@@ -0,0 +1,5 @@
+# flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps
+
+[Fine-tuning settings and fixed asset references](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/vla-finetuning-scope/flappy)
+
+The original checkpoint, configuration and normalization statistics are retained. LoRA sources additionally require their recorded adapter and action-head files.

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps/artifact.json

@@ -0,0 +1,61 @@
+{
+  "run": "flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps",
+  "source": {
+    "repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+    "revision": "1ab96924e691cbe8924ac4671466723bdb5ec3af",
+    "path": "flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps"
+  },
+  "metadata": {
+    "repo": "latency-sensitive-bench/benchmark-models",
+    "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+    "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps"
+  },
+  "checkpoints": [
+    {
+      "repo": "latency-sensitive-bench/benchmark-models",
+      "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+      "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps/checkpoints/steps_3000_model.safetensors"
+    }
+  ],
+  "source_bindings": [
+    {
+      "run": "flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps",
+      "module": "vla-finetuning-scope",
+      "train_latency_raw_frames": 2,
+      "training_step": 3000,
+      "training_started_at_utc": "2026-06-24T09:52:32.813741Z",
+      "git_commit": "68f11e4545a2ad73d3783fd50c14ab886749d0fd",
+      "wandb_run_id": "pllmo179",
+      "checkpoint": "flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps/checkpoints/steps_3000_model.safetensors",
+      "checkpoint_identity": "dee60fc99cf35e2e6d5908129093c8c66c2cafb62d3b87a236d45acd86522d59",
+      "source_path": "flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps/checkpoints/steps_3000_state/model.safetensors",
+      "source_model_revision": "946358fcf7850d259d33e1b822b80cf5aed15577",
+      "freeze_config": {
+        "freeze_modules": "",
+        "freeze_vit": false,
+        "freeze_tied_embedding": false,
+        "freeze_llm_layers": []
+      },
+      "dataset": {
+        "repo": "latency-sensitive-bench/benchmark-datasets",
+        "revision": "e0f4007f504978f93c32557ea4e9b97fdf62fcf5",
+        "path": "latency-aware/flappy/flappy_200ep/flappy_fix_latency_2_200ep",
+        "source_repo": "latency-sensitive-bench/flappy_200ep",
+        "source_revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+        "train_rows": 330734,
+        "train_episodes": 180,
+        "original_source": {
+          "repo": "latency-sensitive-bench/flappy_200ep",
+          "revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+          "path": "flappy_fix_latency_2_200ep"
+        }
+      },
+      "paper_consumers": []
+    }
+  ],
+  "evaluation": {
+    "repo": "latency-sensitive-bench/benchmark-datasets",
+    "revision": "58157c572dbd2e0e5036ce197c42db975e56ae4e",
+    "path": "evaluations/flappy/flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_3k_steps"
+  }
+}

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps/README.md

@@ -0,0 +1,5 @@
+# flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps
+
+[Fine-tuning settings and fixed asset references](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/vla-finetuning-scope/flappy)
+
+The original checkpoint, configuration and normalization statistics are retained. LoRA sources additionally require their recorded adapter and action-head files.

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps/artifact.json

@@ -0,0 +1,61 @@
+{
+  "run": "flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps",
+  "source": {
+    "repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+    "revision": "1ab96924e691cbe8924ac4671466723bdb5ec3af",
+    "path": "flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps"
+  },
+  "metadata": {
+    "repo": "latency-sensitive-bench/benchmark-models",
+    "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+    "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps"
+  },
+  "checkpoints": [
+    {
+      "repo": "latency-sensitive-bench/benchmark-models",
+      "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+      "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps/checkpoints/steps_4000_model.safetensors"
+    }
+  ],
+  "source_bindings": [
+    {
+      "run": "flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps",
+      "module": "vla-finetuning-scope",
+      "train_latency_raw_frames": 2,
+      "training_step": 4000,
+      "training_started_at_utc": "2026-06-24T11:59:06.297366Z",
+      "git_commit": "7b77b87cb564e33377cc2a4f17fc513677849877",
+      "wandb_run_id": "w9uahhsz",
+      "checkpoint": "flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps/checkpoints/steps_4000_model.safetensors",
+      "checkpoint_identity": "074a513f33ae5b81499478f5b7cd258757d271ce29077b30e299b49f5ba81bdb",
+      "source_path": "flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps/checkpoints/steps_4000_state/model.safetensors",
+      "source_model_revision": "946358fcf7850d259d33e1b822b80cf5aed15577",
+      "freeze_config": {
+        "freeze_modules": "",
+        "freeze_vit": false,
+        "freeze_tied_embedding": false,
+        "freeze_llm_layers": []
+      },
+      "dataset": {
+        "repo": "latency-sensitive-bench/benchmark-datasets",
+        "revision": "e0f4007f504978f93c32557ea4e9b97fdf62fcf5",
+        "path": "latency-aware/flappy/flappy_200ep/flappy_fix_latency_2_200ep",
+        "source_repo": "latency-sensitive-bench/flappy_200ep",
+        "source_revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+        "train_rows": 330734,
+        "train_episodes": 180,
+        "original_source": {
+          "repo": "latency-sensitive-bench/flappy_200ep",
+          "revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+          "path": "flappy_fix_latency_2_200ep"
+        }
+      },
+      "paper_consumers": []
+    }
+  ],
+  "evaluation": {
+    "repo": "latency-sensitive-bench/benchmark-datasets",
+    "revision": "58157c572dbd2e0e5036ce197c42db975e56ae4e",
+    "path": "evaluations/flappy/flappy_fix_latency_2_200ep_full_tuning_corrected_bs32_4k_steps"
+  }
+}

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers/README.md

@@ -0,0 +1,5 @@
+# flappy_fix_latency_2_200ep_last_8_layers
+
+[Fine-tuning settings and fixed asset references](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/vla-finetuning-scope/flappy)
+
+The original checkpoint, configuration and normalization statistics are retained. LoRA sources additionally require their recorded adapter and action-head files.

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers/artifact.json

@@ -0,0 +1,61 @@
+{
+  "run": "flappy_fix_latency_2_200ep_last_8_layers",
+  "source": {
+    "repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+    "revision": "1ab96924e691cbe8924ac4671466723bdb5ec3af",
+    "path": "flappy_fix_latency_2_200ep_last_8_layers"
+  },
+  "metadata": {
+    "repo": "latency-sensitive-bench/benchmark-models",
+    "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+    "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers"
+  },
+  "checkpoints": [
+    {
+      "repo": "latency-sensitive-bench/benchmark-models",
+      "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+      "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers/checkpoints/steps_5000_model.safetensors"
+    }
+  ],
+  "source_bindings": [
+    {
+      "run": "flappy_fix_latency_2_200ep_last_8_layers",
+      "module": "vla-finetuning-scope",
+      "train_latency_raw_frames": 2,
+      "training_step": 5000,
+      "training_started_at_utc": "2026-06-23T03:14:55.765555Z",
+      "git_commit": "daeb076ebed09fb21a1a1b3ce6ab124c64018212",
+      "wandb_run_id": "kbm7xajn",
+      "checkpoint": "flappy_fix_latency_2_200ep_last_8_layers/checkpoints/steps_5000_model.safetensors",
+      "checkpoint_identity": "8afaa32fca8bd02a62fd251a0457e809370c64803c90166b0962241a2089519a",
+      "source_path": "flappy_fix_latency_2_200ep_last_8_layers/checkpoints/steps_5000_state/model.safetensors",
+      "source_model_revision": "946358fcf7850d259d33e1b822b80cf5aed15577",
+      "freeze_config": {
+        "freeze_modules": "",
+        "freeze_llm_bottom_ratio": 0.7778
+      },
+      "dataset": {
+        "repo": "latency-sensitive-bench/benchmark-datasets",
+        "revision": "e0f4007f504978f93c32557ea4e9b97fdf62fcf5",
+        "path": "latency-aware/flappy/flappy_200ep/flappy_fix_latency_2_200ep",
+        "source_repo": "latency-sensitive-bench/flappy_200ep",
+        "source_revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+        "train_rows": 330734,
+        "train_episodes": 180,
+        "original_source": {
+          "repo": "latency-sensitive-bench/flappy_200ep",
+          "revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+          "path": "flappy_fix_latency_2_200ep"
+        }
+      },
+      "paper_consumers": [
+        "legacy/vla_hyper"
+      ]
+    }
+  ],
+  "evaluation": {
+    "repo": "latency-sensitive-bench/benchmark-datasets",
+    "revision": "58157c572dbd2e0e5036ce197c42db975e56ae4e",
+    "path": "evaluations/flappy/flappy_fix_latency_2_200ep_last_8_layers"
+  }
+}

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers_corrected/README.md

@@ -0,0 +1,5 @@
+# flappy_fix_latency_2_200ep_last_8_layers_corrected
+
+[Fine-tuning settings and fixed asset references](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/vla-finetuning-scope/flappy)
+
+The original checkpoint, configuration and normalization statistics are retained. LoRA sources additionally require their recorded adapter and action-head files.

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers_corrected/artifact.json

@@ -0,0 +1,61 @@
+{
+  "run": "flappy_fix_latency_2_200ep_last_8_layers_corrected",
+  "source": {
+    "repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+    "revision": "1ab96924e691cbe8924ac4671466723bdb5ec3af",
+    "path": "flappy_fix_latency_2_200ep_last_8_layers_corrected"
+  },
+  "metadata": {
+    "repo": "latency-sensitive-bench/benchmark-models",
+    "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+    "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers_corrected"
+  },
+  "checkpoints": [
+    {
+      "repo": "latency-sensitive-bench/benchmark-models",
+      "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+      "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/steps_5000_model.safetensors"
+    }
+  ],
+  "source_bindings": [
+    {
+      "run": "flappy_fix_latency_2_200ep_last_8_layers_corrected",
+      "module": "vla-finetuning-scope",
+      "train_latency_raw_frames": 2,
+      "training_step": 5000,
+      "training_started_at_utc": "2026-06-23T10:13:20.354969Z",
+      "git_commit": "b6f4f971098c0bc5e340ef616b823a4ca769c7e4",
+      "wandb_run_id": "fomfbmpo",
+      "checkpoint": "flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/steps_5000_model.safetensors",
+      "checkpoint_identity": "750786e2eac8b455ab6e8232416d692e8ad1d5b7a56d386b40bf53a9462ba436",
+      "source_path": "flappy_fix_latency_2_200ep_last_8_layers_corrected/checkpoints/steps_5000_state/model.safetensors",
+      "source_model_revision": "946358fcf7850d259d33e1b822b80cf5aed15577",
+      "freeze_config": {
+        "freeze_modules": "",
+        "freeze_llm_bottom_ratio": 0.7778
+      },
+      "dataset": {
+        "repo": "latency-sensitive-bench/benchmark-datasets",
+        "revision": "e0f4007f504978f93c32557ea4e9b97fdf62fcf5",
+        "path": "latency-aware/flappy/flappy_200ep/flappy_fix_latency_2_200ep",
+        "source_repo": "latency-sensitive-bench/flappy_200ep",
+        "source_revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+        "train_rows": 330734,
+        "train_episodes": 180,
+        "original_source": {
+          "repo": "latency-sensitive-bench/flappy_200ep",
+          "revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+          "path": "flappy_fix_latency_2_200ep"
+        }
+      },
+      "paper_consumers": [
+        "exps/vla_finetuning_scope"
+      ]
+    }
+  ],
+  "evaluation": {
+    "repo": "latency-sensitive-bench/benchmark-datasets",
+    "revision": "58157c572dbd2e0e5036ce197c42db975e56ae4e",
+    "path": "evaluations/flappy/flappy_fix_latency_2_200ep_last_8_layers_corrected"
+  }
+}

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers_corrected_v2/README.md

@@ -0,0 +1,5 @@
+# flappy_fix_latency_2_200ep_last_8_layers_corrected_v2
+
+[Fine-tuning settings and fixed asset references](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/vla-finetuning-scope/flappy)
+
+The original checkpoint, configuration and normalization statistics are retained. LoRA sources additionally require their recorded adapter and action-head files.

latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers_corrected_v2/artifact.json

@@ -0,0 +1,66 @@
+{
+  "run": "flappy_fix_latency_2_200ep_last_8_layers_corrected_v2",
+  "source": {
+    "repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+    "revision": "1ab96924e691cbe8924ac4671466723bdb5ec3af",
+    "path": "flappy_fix_latency_2_200ep_last_8_layers_corrected_v2"
+  },
+  "metadata": {
+    "repo": "latency-sensitive-bench/benchmark-models",
+    "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+    "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers_corrected_v2"
+  },
+  "checkpoints": [
+    {
+      "repo": "latency-sensitive-bench/benchmark-models",
+      "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+      "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/flappy_fix_latency_2_200ep_last_8_layers_corrected_v2/checkpoints/steps_5000_model.safetensors"
+    }
+  ],
+  "source_bindings": [
+    {
+      "run": "flappy_fix_latency_2_200ep_last_8_layers_corrected_v2",
+      "module": "vla-finetuning-scope",
+      "train_latency_raw_frames": 2,
+      "training_step": 5000,
+      "training_started_at_utc": "2026-06-23T16:43:48.375941Z",
+      "git_commit": "fcb51e2694288bdcf8dc8f558e9973cdaaafb944",
+      "wandb_run_id": "z0dfovtb",
+      "checkpoint": "flappy_fix_latency_2_200ep_last_8_layers_corrected_v2/checkpoints/steps_5000_model.safetensors",
+      "checkpoint_identity": "b2375e35dca20389af04508aba967acffda126a63d3daa8344629d271bdcac3d",
+      "source_path": "flappy_fix_latency_2_200ep_last_8_layers_corrected_v2/checkpoints/steps_5000_state/model.safetensors",
+      "source_model_revision": "946358fcf7850d259d33e1b822b80cf5aed15577",
+      "freeze_config": {
+        "freeze_modules": "",
+        "freeze_vit": true,
+        "freeze_tied_embedding": true,
+        "freeze_llm_layers": [
+          0,
+          27
+        ]
+      },
+      "dataset": {
+        "repo": "latency-sensitive-bench/benchmark-datasets",
+        "revision": "e0f4007f504978f93c32557ea4e9b97fdf62fcf5",
+        "path": "latency-aware/flappy/flappy_200ep/flappy_fix_latency_2_200ep",
+        "source_repo": "latency-sensitive-bench/flappy_200ep",
+        "source_revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+        "train_rows": 330734,
+        "train_episodes": 180,
+        "original_source": {
+          "repo": "latency-sensitive-bench/flappy_200ep",
+          "revision": "5df5688a961dc2c91090155c719bff4a5b248235",
+          "path": "flappy_fix_latency_2_200ep"
+        }
+      },
+      "paper_consumers": [
+        "exps/vla_finetuning_scope"
+      ]
+    }
+  ],
+  "evaluation": {
+    "repo": "latency-sensitive-bench/benchmark-datasets",
+    "revision": "58157c572dbd2e0e5036ce197c42db975e56ae4e",
+    "path": "evaluations/flappy/flappy_fix_latency_2_200ep_last_8_layers_corrected_v2"
+  }
+}

latency-aware/flappy/vla/starvla-qwenoft-h1/openvla_flappy_l2_full/README.md

@@ -0,0 +1,5 @@
+# openvla_flappy_l2_full
+
+[Fine-tuning settings and fixed asset references](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/vla-finetuning-scope/archive/flappy)
+
+The original checkpoint, configuration and normalization statistics are retained. LoRA sources additionally require their recorded adapter and action-head files.

latency-aware/flappy/vla/starvla-qwenoft-h1/openvla_flappy_l2_full/artifact.json

@@ -0,0 +1,78 @@
+{
+  "run": "openvla_flappy_l2_full",
+  "source": {
+    "repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+    "revision": "1ab96924e691cbe8924ac4671466723bdb5ec3af",
+    "path": "historical/openvla_flappy_l2_full"
+  },
+  "metadata": {
+    "repo": "latency-sensitive-bench/benchmark-models",
+    "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+    "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/openvla_flappy_l2_full"
+  },
+  "checkpoints": [
+    {
+      "repo": "latency-sensitive-bench/benchmark-models",
+      "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+      "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/openvla_flappy_l2_full/final_model/pytorch_model.pt"
+    }
+  ],
+  "source_bindings": [
+    {
+      "source_repo": "latency-sensitive-bench/openvla_flappy_l2_full",
+      "source_revision": "2e8e58c6af1ee6f8a3701388b2552a4fa65a041c",
+      "target_repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+      "target_run": "historical/openvla_flappy_l2_full",
+      "artifact_state": "complete_checkpoint_bundle",
+      "framework": "QwenOFT",
+      "run_id": "openvla_flappy_l2_full",
+      "training_step": 2000,
+      "seed": 42,
+      "training_data_root": "/root/latency-sensitive-bench/playground/Datasets/rl_games",
+      "training_data_mix": "flappy_fixed_l2_fs4",
+      "dataset_reference": null,
+      "original_dataset_config": null,
+      "weights": [
+        {
+          "path": "historical/openvla_flappy_l2_full/final_model/pytorch_model.pt",
+          "identity": "f6be353e457ac9cf6dbda6cd2e93692d797016100b69ecea4bcab0ececb9a076",
+          "source_path": "final_model/pytorch_model.pt",
+          "size": 9805972271
+        }
+      ],
+      "paper_locations": [
+        "paper/legacy/figures/sft_comparisons/lora_vs_full"
+      ],
+      "paper_identity_confirmed": false,
+      "interpretation": "The legacy experiment and training models are documented. The paper 100-episode raw evaluation and its exact weight binding were not recovered. The separate HF 50-episode record is preserved unchanged.",
+      "unpublished_training_data": "The recorded local training directory has no recovered exact HF source. No replacement dataset is assumed.",
+      "paper_evaluation_record_difference": [
+        {
+          "method": "full",
+          "hf_episode_count": 50,
+          "hf_mean_return": 18.518000243455173,
+          "paper_episode_count": 100,
+          "paper_mean_return": 22.308000287115576,
+          "same_evaluation_record": false,
+          "hf_timestamp": "2026-06-02T23:20:12.557796+00:00",
+          "note": "The HF 50-episode record and paper 100-episode aggregate are retained separately. The paper local per-episode source files were not recovered."
+        },
+        {
+          "method": "lora",
+          "hf_episode_count": 50,
+          "hf_mean_return": 3.8620000724494457,
+          "paper_episode_count": 100,
+          "paper_mean_return": 3.4820000667870046,
+          "same_evaluation_record": false,
+          "hf_timestamp": "2026-06-02T22:57:10.859128+00:00",
+          "note": "The HF 50-episode record and paper 100-episode aggregate are retained separately. The paper local per-episode source files were not recovered."
+        }
+      ]
+    }
+  ],
+  "evaluation": {
+    "repo": "latency-sensitive-bench/benchmark-datasets",
+    "revision": "58157c572dbd2e0e5036ce197c42db975e56ae4e",
+    "path": "evaluations/flappy/openvla_flappy_l2_full"
+  }
+}

latency-aware/flappy/vla/starvla-qwenoft-h1/openvla_flappy_l2_lora/README.md

@@ -0,0 +1,5 @@
+# openvla_flappy_l2_lora
+
+[Fine-tuning settings and fixed asset references](https://huggingface.co/datasets/latency-sensitive-bench/benchmark-datasets/tree/main/experiments/vla-finetuning-scope/archive/flappy)
+
+The original checkpoint, configuration and normalization statistics are retained. LoRA sources additionally require their recorded adapter and action-head files.

latency-aware/flappy/vla/starvla-qwenoft-h1/openvla_flappy_l2_lora/artifact.json

@@ -0,0 +1,89 @@
+{
+  "run": "openvla_flappy_l2_lora",
+  "source": {
+    "repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+    "revision": "1ab96924e691cbe8924ac4671466723bdb5ec3af",
+    "path": "historical/openvla_flappy_l2_lora"
+  },
+  "metadata": {
+    "repo": "latency-sensitive-bench/benchmark-models",
+    "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+    "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/openvla_flappy_l2_lora"
+  },
+  "checkpoints": [
+    {
+      "repo": "latency-sensitive-bench/benchmark-models",
+      "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+      "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/openvla_flappy_l2_lora/final_model/action_model.pt"
+    },
+    {
+      "repo": "latency-sensitive-bench/benchmark-models",
+      "revision": "f99bd515da2a2a364198848b7f3c2dcaaf9566dd",
+      "path": "latency-aware/flappy/vla/starvla-qwenoft-h1/openvla_flappy_l2_lora/final_model/adapter_model.safetensors"
+    }
+  ],
+  "source_bindings": [
+    {
+      "source_repo": "latency-sensitive-bench/openvla_flappy_l2_lora",
+      "source_revision": "427f4c0582d6eb9419fa39f5942db1d59585993c",
+      "target_repo": "latency-sensitive-bench/vla-finetuning-scope-models",
+      "target_run": "historical/openvla_flappy_l2_lora",
+      "artifact_state": "complete_checkpoint_bundle",
+      "framework": "QwenOFT",
+      "run_id": "openvla_flappy_l2_lora",
+      "training_step": 2000,
+      "seed": 42,
+      "training_data_root": "/mnt/data/latency_data/playground/Datasets/rl_games",
+      "training_data_mix": "flappy_fixed_l2_fs4",
+      "dataset_reference": null,
+      "original_dataset_config": null,
+      "weights": [
+        {
+          "path": "historical/openvla_flappy_l2_lora/final_model/action_model.pt",
+          "identity": "e8bc83aaf046d0b6c2045cd38c06875ccc60c808701d6eba10fc491e2b016ffb",
+          "source_path": "final_model/action_model.pt",
+          "size": 131200330
+        },
+        {
+          "path": "historical/openvla_flappy_l2_lora/final_model/adapter_model.safetensors",
+          "identity": "551bd2362cc9548b9d97f3a4f77fcfbe2b28f2725b2897ab42b996485962b8bb",
+          "source_path": "final_model/adapter_model.safetensors",
+          "size": 161192096
+        }
+      ],
+      "paper_locations": [
+        "paper/legacy/figures/sft_comparisons/lora_vs_full"
+      ],
+      "paper_identity_confirmed": false,
+      "interpretation": "The legacy experiment and training models are documented. The paper 100-episode raw evaluation and its exact weight binding were not recovered. The separate HF 50-episode record is preserved unchanged.",
+      "unpublished_training_data": "The recorded local training directory has no recovered exact HF source. No replacement dataset is assumed.",
+      "paper_evaluation_record_difference": [
+        {
+          "method": "full",
+          "hf_episode_count": 50,
+          "hf_mean_return": 18.518000243455173,
+          "paper_episode_count": 100,
+          "paper_mean_return": 22.308000287115576,
+          "same_evaluation_record": false,
+          "hf_timestamp": "2026-06-02T23:20:12.557796+00:00",
+          "note": "The HF 50-episode record and paper 100-episode aggregate are retained separately. The paper local per-episode source files were not recovered."
+        },
+        {
+          "method": "lora",
+          "hf_episode_count": 50,
+          "hf_mean_return": 3.8620000724494457,
+          "paper_episode_count": 100,
+          "paper_mean_return": 3.4820000667870046,
+          "same_evaluation_record": false,
+          "hf_timestamp": "2026-06-02T22:57:10.859128+00:00",
+          "note": "The HF 50-episode record and paper 100-episode aggregate are retained separately. The paper local per-episode source files were not recovered."
+        }
+      ]
+    }
+  ],
+  "evaluation": {
+    "repo": "latency-sensitive-bench/benchmark-datasets",
+    "revision": "58157c572dbd2e0e5036ce197c42db975e56ae4e",
+    "path": "evaluations/flappy/openvla_flappy_l2_lora"
+  }
+}