Open-SWE-Traces

Open-SWE-Traces: Advancing Distillation for Software Engineering Agents Data Overview Open-SWE-Traces is an agentic instruction tuning dataset designed to advance the capabilities of LLMs in software engineering. This dataset comprises 200k+ agent trajectories collected using the SWE-agent and OpenHands framework. The trajectories were synthesized using Minimax-M2.5 (with thinking) and Qwen3.5-122B-A10B (without thinking) and specifically curated for supervised fine-tuning (SFT), aiming to improve model performance on SWE-Bench style tasks. The issue statements are sourced from SWE-rebench-V2 with permissive licenses (MIT, Apache-2.0, BSD-2-Clause, BSD-3-Clause). This dataset is ready for commercial/non-commercial use. Data Fields |Field|Type|Description| |:---|:---|:---| | `instance_id` | `string`| Unique identifier for the instance (from nebius/SWE-rebench-V2)| | `repo` | `string` | GitHub repository in `owner/repo` format| | `license` | `string` | SPDX license identifier of the repository| | `language` | `string` | Primary programming language of the repository.| | `trajectory_id` | `string` | The identifier unique for each collected trajectory.| | `trajectory` | `list` | Complete conversation history with roles: 'system' and 'user' (initial prompts), 'assistant' (model reasoning/actions) and 'tool' (environment observations).| | `tools` | `list[string]` | A list of tool definitions available to the agent during execution, serialized as JSON strings.| | `resolved` | `int` | Indicator of task success: 1 if the agent solved the issue, 0 otherwise. (-1 if the information is unavailable)| | `metadata` | `dict` | Additional contextual info about the PR task and execution metrics: • **`category`** (`string`): The issue classification type (e.g., `"bug-fix"`, `"enhancement"`). • **`reference_patch`** (`dict`): Reference ground-truth patch details:     — *`patch`* (`string`): The gold patch in unified diff format.     — *`num_modif