Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456 Paper page - SWE-Lego: Pushing the Limits of Supervised Fine-tuning for Software Issue Resolving
\n","updatedAt":"2026-01-06T09:02:56.749Z","author":{"_id":"63c20105726f62e411fbe882","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63c20105726f62e411fbe882/2UsU9O2psbDjJzz-sAmGH.jpeg","fullname":"Yuxin Jiang","name":"YuxinJiang","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":6,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.4301125705242157},"editors":["YuxinJiang"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/63c20105726f62e411fbe882/2UsU9O2psbDjJzz-sAmGH.jpeg"],"reactions":[],"isReport":false}},{"id":"695db8fdb9a85af4c60c104e","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false},"createdAt":"2026-01-07T01:38:05.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [Training Versatile Coding Agents in Synthetic Environments](https://huggingface.co/papers/2512.12216) (2025)\n* [SWE-EVO: Benchmarking Coding Agents in Long-Horizon Software Evolution Scenarios](https://huggingface.co/papers/2512.18470) (2025)\n* [SWE-RM: Execution-free Feedback For Software Engineering Agents](https://huggingface.co/papers/2512.21919) (2025)\n* [Klear-AgentForge: Forging Agentic Intelligence through Posttraining Scaling](https://huggingface.co/papers/2511.05951) (2025)\n* [One Tool Is Enough: Reinforcement Learning for Repository-Level LLM Agents](https://huggingface.co/papers/2512.20957) (2025)\n* [Nanbeige4-3B Technical Report: Exploring the Frontier of Small Language Models](https://huggingface.co/papers/2512.06266) (2025)\n* [OpenMMReasoner: Pushing the Frontiers for Multimodal Reasoning with an Open and General Recipe](https://huggingface.co/papers/2511.16334) (2025)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
\n
The following papers were recommended by the Semantic Scholar API
Please give a thumbs up to this comment if you found it helpful!
\n
If you want recommendations for any Paper on Hugging Face checkout this Space
\n
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend
\n","updatedAt":"2026-01-07T01:38:05.037Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6957316994667053},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}},{"id":"696be570ca5779cca933999b","author":{"_id":"65243980050781c16f234f1f","avatarUrl":"/avatars/743a009681d5d554c27e04300db9f267.svg","fullname":"Avi","name":"avahal","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false},"createdAt":"2026-01-17T19:39:28.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"arXivlens breakdown of this paper 👉 https://arxivlens.com/PaperView/Details/swe-lego-pushing-the-limits-of-supervised-fine-tuning-for-software-issue-resolving-5760-dcca0e8e\n\n- Executive Summary\n- Detailed Breakdown\n- Practical Applications","html":"
\n","updatedAt":"2026-01-17T19:39:28.618Z","author":{"_id":"65243980050781c16f234f1f","avatarUrl":"/avatars/743a009681d5d554c27e04300db9f267.svg","fullname":"Avi","name":"avahal","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7375622987747192},"editors":["avahal"],"editorAvatarUrls":["/avatars/743a009681d5d554c27e04300db9f267.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2601.01426","authors":[{"_id":"695c7a8a6aa73bc11f091411","user":{"_id":"60fc2fcca6bdebbe52dfdaf4","avatarUrl":"/avatars/1d59a7f33cb0df04678516f337e6b881.svg","isPro":false,"fullname":"Chaofan Tao","user":"tcftrees","type":"user"},"name":"Chaofan Tao","status":"claimed_verified","statusLastChangedAt":"2026-01-07T09:27:37.648Z","hidden":false},{"_id":"695c7a8a6aa73bc11f091412","user":{"_id":"65e0d0f417458030d05b5622","avatarUrl":"/avatars/bb35db227e946bc8dbfcba871f32730a.svg","isPro":false,"fullname":"Jierun Chen","user":"JierunChen","type":"user"},"name":"Jierun Chen","status":"claimed_verified","statusLastChangedAt":"2026-01-07T09:27:31.750Z","hidden":false},{"_id":"695c7a8a6aa73bc11f091413","user":{"_id":"63c20105726f62e411fbe882","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63c20105726f62e411fbe882/2UsU9O2psbDjJzz-sAmGH.jpeg","isPro":false,"fullname":"Yuxin Jiang","user":"YuxinJiang","type":"user"},"name":"Yuxin Jiang","status":"claimed_verified","statusLastChangedAt":"2026-01-06T09:58:05.382Z","hidden":false},{"_id":"695c7a8a6aa73bc11f091414","name":"Kaiqi Kou","hidden":false},{"_id":"695c7a8a6aa73bc11f091415","name":"Shaowei Wang","hidden":false},{"_id":"695c7a8a6aa73bc11f091416","user":{"_id":"6458899acfd4a6e2bc005fd9","avatarUrl":"/avatars/b7b305a8baa3cf7683442e2a650306d6.svg","isPro":false,"fullname":"Ruoyu Wang","user":"ruoyu001","type":"user"},"name":"Ruoyu Wang","status":"claimed_verified","statusLastChangedAt":"2026-01-12T10:36:31.574Z","hidden":false},{"_id":"695c7a8a6aa73bc11f091417","name":"Xiaohui Li","hidden":false},{"_id":"695c7a8a6aa73bc11f091418","name":"Sidi Yang","hidden":false},{"_id":"695c7a8a6aa73bc11f091419","user":{"_id":"68954ff742413a5846d9d80a","avatarUrl":"/avatars/1751b607a0a65ba63400d34487f3b75c.svg","isPro":false,"fullname":"ElvinDu","user":"ElvinDu518","type":"user"},"name":"Yiming Du","status":"claimed_verified","statusLastChangedAt":"2026-01-07T09:27:33.795Z","hidden":false},{"_id":"695c7a8a6aa73bc11f09141a","user":{"_id":"61669c456916c52acd5a1aa3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/61669c456916c52acd5a1aa3/HnZTwRaXgTeTG3ljO3ITb.jpeg","isPro":false,"fullname":"jianbo dai","user":"jbd","type":"user"},"name":"Jianbo Dai","status":"claimed_verified","statusLastChangedAt":"2026-01-09T08:36:42.643Z","hidden":false},{"_id":"695c7a8a6aa73bc11f09141b","name":"Zhiming Mao","hidden":false},{"_id":"695c7a8a6aa73bc11f09141c","name":"Xinyu Wang","hidden":false},{"_id":"695c7a8a6aa73bc11f09141d","name":"Lifeng Shang","hidden":false},{"_id":"695c7a8a6aa73bc11f09141e","name":"Haoli Bai","hidden":false}],"publishedAt":"2026-01-04T08:07:27.000Z","submittedOnDailyAt":"2026-01-06T06:29:01.298Z","title":"SWE-Lego: Pushing the Limits of Supervised Fine-tuning for Software Issue Resolving","submittedOnDailyBy":{"_id":"63c20105726f62e411fbe882","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63c20105726f62e411fbe882/2UsU9O2psbDjJzz-sAmGH.jpeg","isPro":false,"fullname":"Yuxin Jiang","user":"YuxinJiang","type":"user"},"summary":"We present SWE-Lego, a supervised fine-tuning (SFT) recipe designed to achieve state-ofthe-art performance in software engineering (SWE) issue resolving. In contrast to prevalent methods that rely on complex training paradigms (e.g., mid-training, SFT, reinforcement learning, and their combinations), we explore how to push the limits of a lightweight SFT-only approach for SWE tasks. SWE-Lego comprises three core building blocks, with key findings summarized as follows: 1) the SWE-Lego dataset, a collection of 32k highquality task instances and 18k validated trajectories, combining real and synthetic data to complement each other in both quality and quantity; 2) a refined SFT procedure with error masking and a difficulty-based curriculum, which demonstrably improves action quality and overall performance. Empirical results show that with these two building bricks alone,the SFT can push SWE-Lego models to state-of-the-art performance among open-source models of comparable size on SWE-bench Verified: SWE-Lego-Qwen3-8B reaches 42.2%, and SWE-Lego-Qwen3-32B attains 52.6%. 3) We further evaluate and improve test-time scaling (TTS) built upon the SFT foundation. Based on a well-trained verifier, SWE-Lego models can be significantly boosted--for example, 42.2% to 49.6% and 52.6% to 58.8% under TTS@16 for the 8B and 32B models, respectively.","upvotes":24,"discussionId":"695c7a8a6aa73bc11f09141f","projectPage":"https://github.com/SWE-Lego/SWE-Lego","githubRepo":"https://github.com/SWE-Lego/SWE-Lego","githubRepoAddedBy":"user","ai_summary":"SWE-Lego achieves state-of-the-art performance in software engineering task resolution through a lightweight supervised fine-tuning approach combined with a curated dataset and refined training procedures.","ai_keywords":["supervised fine-tuning","SWE-bench Verified","error masking","difficulty-based curriculum","test-time scaling","verifier"],"githubStars":57,"organization":{"_id":"6951e68428a36c372970db39","name":"SWE-Lego","fullname":"SWE-Lego","avatar":"https://cdn-uploads.huggingface.co/production/uploads/60fc2fcca6bdebbe52dfdaf4/AeuYwUH-CQCt893qnmAGa.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"63c20105726f62e411fbe882","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63c20105726f62e411fbe882/2UsU9O2psbDjJzz-sAmGH.jpeg","isPro":false,"fullname":"Yuxin Jiang","user":"YuxinJiang","type":"user"},{"_id":"6458899acfd4a6e2bc005fd9","avatarUrl":"/avatars/b7b305a8baa3cf7683442e2a650306d6.svg","isPro":false,"fullname":"Ruoyu Wang","user":"ruoyu001","type":"user"},{"_id":"667e1a0fa0ffad3dbc60df9d","avatarUrl":"/avatars/e16c7a3620ae5c82b7b636296a975ae1.svg","isPro":false,"fullname":"Haoli Bai","user":"baihaoli","type":"user"},{"_id":"6599556028676374f370807a","avatarUrl":"/avatars/85f1d6c7a6f5f3fd2a8db3fe41d7125f.svg","isPro":false,"fullname":"Joanna","user":"Joannananananana","type":"user"},{"_id":"60fc2fcca6bdebbe52dfdaf4","avatarUrl":"/avatars/1d59a7f33cb0df04678516f337e6b881.svg","isPro":false,"fullname":"Chaofan Tao","user":"tcftrees","type":"user"},{"_id":"655b4f5cc11dee7f7e882a0c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/655b4f5cc11dee7f7e882a0c/4GaAC5Qt55eYlbNRJpVD5.png","isPro":false,"fullname":"zunhaisu","user":"zunhai","type":"user"},{"_id":"63870c8388b39a64e1e8cdfa","avatarUrl":"/avatars/1813b49eca6eb7396fa18cccc6e24342.svg","isPro":false,"fullname":"zhanghengyuan","user":"hengyuanya","type":"user"},{"_id":"63d9020e9dfcfa941d4f6a52","avatarUrl":"/avatars/8e9608ae1a45c3a33f12d84491c2d161.svg","isPro":false,"fullname":"SChen","user":"schen021","type":"user"},{"_id":"6364cc3912188d67e6526cfa","avatarUrl":"/avatars/bb5c3200c6d64d1260b7e418cf1a76f9.svg","isPro":false,"fullname":"Jiacheng Ye","user":"jiacheng-ye","type":"user"},{"_id":"68954ff742413a5846d9d80a","avatarUrl":"/avatars/1751b607a0a65ba63400d34487f3b75c.svg","isPro":false,"fullname":"ElvinDu","user":"ElvinDu518","type":"user"},{"_id":"683412481c06c92ca6d31ea7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/l7Q6gayiZaE2KKn1fgCDk.png","isPro":false,"fullname":"Shuqi Wang","user":"QiqiqiqiWa","type":"user"},{"_id":"65e0d0f417458030d05b5622","avatarUrl":"/avatars/bb35db227e946bc8dbfcba871f32730a.svg","isPro":false,"fullname":"Jierun Chen","user":"JierunChen","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0,"organization":{"_id":"6951e68428a36c372970db39","name":"SWE-Lego","fullname":"SWE-Lego","avatar":"https://cdn-uploads.huggingface.co/production/uploads/60fc2fcca6bdebbe52dfdaf4/AeuYwUH-CQCt893qnmAGa.png"}}">
SWE-Lego achieves state-of-the-art performance in software engineering task resolution through a lightweight supervised fine-tuning approach combined with a curated dataset and refined training procedures.
AI-generated summary
We present SWE-Lego, a supervised fine-tuning (SFT) recipe designed to achieve state-ofthe-art performance in software engineering (SWE) issue resolving. In contrast to prevalent methods that rely on complex training paradigms (e.g., mid-training, SFT, reinforcement learning, and their combinations), we explore how to push the limits of a lightweight SFT-only approach for SWE tasks. SWE-Lego comprises three core building blocks, with key findings summarized as follows: 1) the SWE-Lego dataset, a collection of 32k highquality task instances and 18k validated trajectories, combining real and synthetic data to complement each other in both quality and quantity; 2) a refined SFT procedure with error masking and a difficulty-based curriculum, which demonstrably improves action quality and overall performance. Empirical results show that with these two building bricks alone,the SFT can push SWE-Lego models to state-of-the-art performance among open-source models of comparable size on SWE-bench Verified: SWE-Lego-Qwen3-8B reaches 42.2%, and SWE-Lego-Qwen3-32B attains 52.6%. 3) We further evaluate and improve test-time scaling (TTS) built upon the SFT foundation. Based on a well-trained verifier, SWE-Lego models can be significantly boosted--for example, 42.2% to 49.6% and 52.6% to 58.8% under TTS@16 for the 8B and 32B models, respectively.