SWE-Prime: Fewer Trajectories, Better Performance
2026-08-27 • Software Engineering
Software EngineeringArtificial IntelligenceComputation and Language
AI summaryⓘ
The authors found that just using all successful problem-solving examples to train language models can be harmful because some steps are inefficient or risky. They developed SWE-Prime, a method that carefully picks the best examples by first filtering whole problem-solving attempts and then selecting useful parts of those attempts. This way, models learn from higher-quality, more helpful data. Their tests showed that training on this smartly chosen smaller dataset led to better performance than using all available data.
large language modelssupervised fine-tuningsoftware problem solvingtrajectory datadata selectionsemantic segmentationtraining data qualitymodel performanceSWE-PrimeSWE-Bench
Authors
Dewu Zheng, Ruizhe Ye, Yanlin Wang, Yang Ye, Hongyu Zhang, Ensheng Shi, Xilin Liu, Yuchi Ma, Jianxing Yu, Zibin Zheng
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-Prime, a multi-granularity, two-stage SFT data selection method that progressively filters training data at the trajectory and segment levels. Specifically, the first stage performs trajectory-level screening based on process quality, result quality, and data representativeness, selecting a high-quality and representative subset of successful trajectories. The second stage performs segment-level selection by grouping consecutive steps into semantic segments and assessing each segment based on its contribution to the final solution, learnability, and potential risks. During SFT, all segments remain in the sequence to preserve context, while only selected segments contribute to the loss computation. Experiments on SWE-Bench Pro and SWE-Bench Verified show that training on the 10% trajectory subset selected by SWE-Prime outperforms training on the full resolved dataset, yielding relative performance gains of up to 12.2% and 24.2%, respectively.