swe-lego-issue-resolution

【数据与研究】Achieve state-of-the-art software issue resolution through lightweight supervised fine-tuning with structured datasets and curriculum learning, plus test-time scaling—reaching 52.6% on SWE-Bench Verified and 58.8% with test-time strategies, outperforming complex multi-stage training.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准