Synthetic GitHub Issue Conversations for Agent Training

11,500 synthetic GitHub issues in Python and their multi-turn agent trajectories. The dataset was created by the SWE-Lego project using the Qwen/Qwen3-Coder-480B-A35B-Instruct model with OpenHands agent scaffolding and was last updated in January 2026.

Use Cases

• Supervised fine-tuning of language models using the `messages` column containing multi-turn agent trajectories.
• Training conversational agents to handle Python-related GitHub issue discussions based on synthetic issue content.
• Benchmarking agent performance on code-related problem-solving tasks using the structured conversation trajectories.
• Analyzing patterns in synthetic software engineering dialogues for curriculum learning or data augmentation.

Strengths

• Contains 11,500 synthetic data points for training.
• Includes multi-turn agent trajectories for each synthetic issue.

Limitations

• Data is entirely synthetic, which may not fully capture the complexity and noise of real-world GitHub issues.
• Limited to Python language issues, reducing applicability to other programming ecosystems.
• Unknown sample size and potential class distribution for different issue types.

Provenance

Source

SWE-Lego project, hosted on Hugging Face.

Collection Method

Generated synthetically using the Qwen/Qwen3-Coder-480B-A35B-Instruct model with OpenHands (v0.53.0) agent scaffolding.

Freshness

Last updated January 2026.