timchen0618/bcp-traj-ext-formatted-v1

--- license: mit tags: - bcp - traj-ext - browsecomp-plus - seed0 --- # bcp-traj-ext-formatted-v1 Trajectories from seed0 (gpt-oss-120b, Qwen3-Embedding-8B, full split) formatted in the traj_ext style: trajectory_text is the serialized steps ([Reasoning]/[Tool call]/[Tool result]/[Final answer]), and formatted_prompt is the full QUERY_TEMPLATE_GIVEN_TRAJECTORY prompt ready to feed to the next agent. ## Dataset Info - **Rows**: 830 - **Columns**: 9 ## Columns | Column | Type | Description | |--------|------|-------------| | query_id | Value('string') | BrowseComp-Plus query ID | | question | Value('string') | Raw question text | | trajectory_text | Value('string') | Serialized trajectory: [Reasoning]/[Tool call]/[Tool result]/[Final answer] blocks joined by double newlines. Tool outputs truncated to 500 chars. | | formatted_prompt | Value('string') | Full QUERY_TEMPLATE_GIVEN_TRAJECTORY prompt with question and <trajectory>...</trajectory> block — the exact string fed to the model in traj_ext mode. | | status | Value('string') | Trajectory completion status ('completed' or 'incomplete') | | n_steps | Value('int64') | Total number of result steps in the trajectory | | n_tool_calls | Value('int64') | Number of tool call steps | | n_reasoning_steps | Value('int64') | Number of reasoning steps | | run_id | Value('string') | Source trajectory filename (without .json) | ## Generation Parameters ```json { "script_name": "format_traj_ext.py", "model": "gpt-oss-120b", "description": "Trajectories from seed0 (gpt-oss-120b, Qwen3-Embedding-8B, full split) formatted in the traj_ext style: trajectory_text is the serialized steps ([Reasoning]/[Tool call]/[Tool result]/[Final answer]), and formatted_prompt is the full QUERY_TEMPLATE_GIVEN_TRAJECTORY prompt ready to feed to the next agent.", "cluster": "torch", "artifact_status": "final", "canary": false, "hyperparameters": {}, "input_datasets": [] } ``` ## Usage ```python from datasets import load_dataset dataset = load_dataset("timchen0618/bcp-traj-ext-formatted-v1", split="train") print(f"Loaded {len(dataset)} rows") ``` ---