Upload folder using huggingface_hub

.gitattributes

@@ -58,3 +58,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+task_a_9b/train.jsonl filter=lfs diff=lfs merge=lfs -text
+task_b_9b/train.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,173 @@
+---
+language:
+- en
+license: apache-2.0
+task_categories:
+- text-generation
+pretty_name: BeliefTrack
+configs:
+- config_name: task_a_7b
+  data_files:
+  - split: train
+    path: task_a_7b/train.jsonl
+  - split: test
+    path: task_a_7b/test.jsonl
+- config_name: task_a_9b
+  data_files:
+  - split: train
+    path: task_a_9b/train.jsonl
+  - split: test
+    path: task_a_9b/test.jsonl
+- config_name: task_b_7b
+  data_files:
+  - split: train
+    path: task_b_7b/train.jsonl
+  - split: test
+    path: task_b_7b/test.jsonl
+- config_name: task_b_9b
+  data_files:
+  - split: train
+    path: task_b_9b/train.jsonl
+  - split: test
+    path: task_b_9b/test.jsonl
+---
+
+# BeliefTrack
+
+BeliefTrack is a closed-world benchmark for **Contextual Belief Management (CBM)** in multi-turn language model interactions. Each example asks a model to maintain the set of hypotheses that remain consistent with formal evidence.
+
+## Configurations
+
+| Config | Task | Model Setting | Splits |
+|---|---|---|---|
+| `task_a_7b` | Rule Discovery | 7B experiment setting | train/test |
+| `task_a_9b` | Rule Discovery | 9B thinking experiment setting | train/test |
+| `task_b_7b` | Circuit Diagnosis | 7B experiment setting | train/test |
+| `task_b_9b` | Circuit Diagnosis | 9B thinking experiment setting | train/test |
+
+The `7B` and `9B` names identify the experimental data-generation/evaluation setting used in the project. They do not restrict the dataset to those model sizes.
+
+## Failure Modes
+
+- `failed_stay`: redundant evidence should not change the belief state.
+- `failed_update`: a correction should replace earlier formal evidence and trigger belief recomputation.
+- `failed_isolation`: task-irrelevant noise should not affect the formal-evidence belief state.
+
+## Loading
+
+```python
+from datasets import load_dataset
+
+ds = load_dataset("YOUR_ORG/BeliefTrack", "task_a_7b")
+print(ds["train"][0])
+```
+
+You can also load directly from local files:
+
+```python
+from datasets import load_dataset
+
+ds = load_dataset(
+    "json",
+    data_files={
+        "train": "task_a_7b/train.jsonl",
+        "test": "task_a_7b/test.jsonl",
+    },
+)
+```
+
+## Schema
+
+Each row contains:
+
+- `case_id`: unique example id.
+- `task`: `task_a` or `task_b`.
+- `task_name`: human-readable task name.
+- `model_setting`: `7B` or `9B`.
+- `split`: `train` or `test`.
+- `cbm_challenge_type`: `failed_stay`, `failed_update`, or `failed_isolation`.
+- `system_prompt`: system instruction shown to the model.
+- `turns_json`: JSON-encoded multi-turn formal-evidence trajectory with per-turn gold belief states when available.
+- `messages_json`: JSON-encoded training-format messages when available.
+- `target_set_json`, `oracle`, `gt_survivors_json`: symbolic verifier targets and labels when available.
+
+## Dataset Summary
+
+```json
+{
+  "configs": {
+    "task_a_7b": {
+      "train": {
+        "source": "data/belief_training_task_dataset/Task_A/7B/train/train_cases_7B.json",
+        "num_rows": 1000,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 500,
+          "failed_update": 500
+        }
+      },
+      "test": {
+        "source": "data/belief_training_task_dataset/Task_A/7B/test/test_cases_7B.json",
+        "num_rows": 200,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 100,
+          "failed_update": 100
+        }
+      }
+    },
+    "task_a_9b": {
+      "train": {
+        "source": "data/belief_training_task_dataset/Task_A/9B/train/train_cases_9B_thinking.json",
+        "num_rows": 1260,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 760,
+          "failed_update": 500
+        }
+      },
+      "test": {
+        "source": "data/belief_training_task_dataset/Task_A/9B/test/test_cases_9B_thinking.json",
+        "num_rows": 200,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 100,
+          "failed_update": 100
+        }
+      }
+    },
+    "task_b_7b": {
+      "train": {
+        "source": "data/belief_training_task_dataset/Task_B/7B/train/train_cases_7B.json",
+        "num_rows": 750,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 200,
+          "failed_update": 550
+        }
+      },
+      "test": {
+        "source": "data/belief_training_task_dataset/Task_B/7B/test/test_cases_7B.json",
+        "num_rows": 199,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 100,
+          "failed_update": 99
+        }
+      }
+    },
+    "task_b_9b": {
+      "train": {
+        "source": "data/belief_training_task_dataset/Task_B/9B/train/train_cases_9B_thinking.json",
+        "num_rows": 1247,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 598,
+          "failed_update": 649
+        }
+      },
+      "test": {
+        "source": "data/belief_training_task_dataset/Task_B/9B/test/test_cases_9B_thinking.json",
+        "num_rows": 326,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 200,
+          "failed_update": 126
+        }
+      }
+    }
+  }
+}
+```

dataset_summary.json

@@ -0,0 +1,76 @@
+{
+  "configs": {
+    "task_a_7b": {
+      "train": {
+        "source": "data/belief_training_task_dataset/Task_A/7B/train/train_cases_7B.json",
+        "num_rows": 1000,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 500,
+          "failed_update": 500
+        }
+      },
+      "test": {
+        "source": "data/belief_training_task_dataset/Task_A/7B/test/test_cases_7B.json",
+        "num_rows": 200,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 100,
+          "failed_update": 100
+        }
+      }
+    },
+    "task_a_9b": {
+      "train": {
+        "source": "data/belief_training_task_dataset/Task_A/9B/train/train_cases_9B_thinking.json",
+        "num_rows": 1260,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 760,
+          "failed_update": 500
+        }
+      },
+      "test": {
+        "source": "data/belief_training_task_dataset/Task_A/9B/test/test_cases_9B_thinking.json",
+        "num_rows": 200,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 100,
+          "failed_update": 100
+        }
+      }
+    },
+    "task_b_7b": {
+      "train": {
+        "source": "data/belief_training_task_dataset/Task_B/7B/train/train_cases_7B.json",
+        "num_rows": 750,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 200,
+          "failed_update": 550
+        }
+      },
+      "test": {
+        "source": "data/belief_training_task_dataset/Task_B/7B/test/test_cases_7B.json",
+        "num_rows": 199,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 100,
+          "failed_update": 99
+        }
+      }
+    },
+    "task_b_9b": {
+      "train": {
+        "source": "data/belief_training_task_dataset/Task_B/9B/train/train_cases_9B_thinking.json",
+        "num_rows": 1247,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 598,
+          "failed_update": 649
+        }
+      },
+      "test": {
+        "source": "data/belief_training_task_dataset/Task_B/9B/test/test_cases_9B_thinking.json",
+        "num_rows": 326,
+        "counts_by_cbm_challenge_type": {
+          "failed_stay": 200,
+          "failed_update": 126
+        }
+      }
+    }
+  }
+}

task_a_9b/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1a30329a5de5c6d83637b9e68dc46cfa38a4c3c1be66d1fa47b252e40efa1a6
+size 42231351

task_b_9b/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4afd36ce89cb019fa2900f9980f522194a4867c007f86205ffa680e450cf8c04
+size 37392842