SWE-Prime: Fewer Trajectories, Better Performance
Quick summary
arXiv:2608.27449v1 Announce Type: cross Abstract: To improve large language models' ability to resolve real-world software issues, prior work has focused on constructing large-scale agent trajectory datasets and performing supervised fine-tuning (SFT) on successful trajectories. However, task success does not guarantee high-quality supervision: successful trajectories may still contain ineffective, redundant, or risky steps. Directly using such trajectories for SFT can introduce noisy supervision and encourage models to imitate undesirable problem-solving behaviors. Therefore, we propose SWE-P
Key takeaways
- arXiv:2608.27449v1 Announce Type: cross Abstract: To improve large language models' ability to resolve real-world software issues, prior work has focused on constructing large-scale agent trajectory datasets and performing supervised fine-tuning (SFT) on successful trajectories.
- However, task success does not guarantee high-quality supervision: successful trajectories may still contain ineffective, redundant, or risky steps.
- Directly using such trajectories for SFT can introduce noisy supervision and encourage models to imitate undesirable problem-solving behaviors.
Why it matters
The importance of “SWE-Prime: Fewer Trajectories, Better Performance” will be measured by what changes in practice. User behavior, access conditions, verifiable performance and responsible-use outcomes are the signals worth following.
Member comments