SWE-Next SFT Trajectories is the supervised fine-tuning dataset released with SWE-Next: Scalable Real-World Software Engineering Tasks for Agents. It contains 3,693 ShareGPT-style multi-turn training examples collected from expert agent rollouts on 2,308 execution-grounded SWE tasks synthesized from real merged pull requests.
The dataset is designed for training… See the full description on the dataset page:
https://huggingface.co/datasets/TIGER-Lab/SWE-Next-SFT-Trajectories.
