[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-SWE-bench--SWE-smith":3,"tool-SWE-bench--SWE-smith":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":75,"owner_avatar_url":76,"owner_bio":77,"owner_company":78,"owner_location":78,"owner_email":78,"owner_twitter":78,"owner_website":79,"owner_url":80,"languages":81,"stars":120,"forks":121,"last_commit_at":122,"license":123,"difficulty_score":10,"env_os":124,"env_gpu":125,"env_ram":125,"env_deps":126,"category_tags":131,"github_topics":132,"view_count":10,"oss_zip_url":78,"oss_zip_packed_at":78,"status":16,"created_at":137,"updated_at":138,"faqs":139,"releases":168},843,"SWE-bench\u002FSWE-smith","SWE-smith","[NeurIPS 2025 D&B Spotlight] Scaling Data for SWE-agents","SWE-smith 是一款面向软件工程师智能体（SWE-agents）训练的开源工具包，旨在解决 AI 编程领域高质量训练数据稀缺的难题。它将任意 GitHub 仓库转化为可交互的“训练场”，能够自动生成包括文件定位、程序修复在内的无限任务实例，让模型在真实代码环境中学习。\n\nSWE-smith 特别适合 AI 算法研究员、后端开发工程师以及希望探索自动化代码修复技术的团队。借助它，用户不仅能构建专属的 SWE-gym 环境，还能直接利用其提供的 5.2 万任务实例数据集进行模型微调。例如，基于此框架微调的 SWE-agent-LM-32B 已在 SWE-bench 验证集中取得了 40.2% 的通过率。\n\n作为 NeurIPS 2025 数据集与基准测试轨道的焦点项目，SWE-smith 采用 MIT 协议开源，支持 Python 3.10+ 及 Docker 环境。虽然目前主要针对 Linux 系统优化，但它为提升大语言模型的软件工程能力提供了强大的基础设施和数据支撑，是构建下一代 AI 程序员的重要基石。","\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fswesmith.com\u002F\">\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FSWE-bench_SWE-smith_readme_a9684cb455f4.png\" style=\"height: 10em\" alt=\"Kawhi the SWE-smith\" \u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\u003Cbr>\n\u003Cdiv align=\"center\">\n\u003Cstrong>NeurIPS 2025 Datasets & Benchmarks Track - Spotlight 🔦\u003C\u002Fstrong>\n\u003C\u002Fdiv>\n\u003Cbr>\n\u003Cdiv align=\"center\">\n\u003Ca href=\"https:\u002F\u002Fwww.python.org\u002F\">\n  \u003Cimg alt=\"Build\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.10+-1f425f.svg?color=purple\">\n\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fcopyright.princeton.edu\u002Fpolicy\">\n  \u003Cimg alt=\"License\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-blue\">\n\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fswesmith\">\n  \u003Cimg src=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fswesmith.svg\">\n\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.21798\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FarXiv-2504.21798-b31b1b.svg\">\n\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Chr \u002F>\n\nSWE-smith is a toolkit for training [SWE-agents](https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FSWE-agent). You can:\n* Turn any Github repository into a [SWE-gym](https:\u002F\u002Fgithub.com\u002FSWE-Gym\u002FSWE-Gym).\n* Create *unlimited* tasks (e.g., file localization, program repair, [SWE-bench](https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-bench)) for that repo.\n* Train an LM to become a better SWE ([SWE-agent-LM-32B](https:\u002F\u002Fhuggingface.co\u002FSWE-bench\u002FSWE-agent-LM-32B)).\n\n## ⚒️ Build Environments\nIf you're interested in turning a GitHub repository into a SWE-gym, install the package from [source](https:\u002F\u002Fswesmith.com\u002Fgetting_started\u002Finstallation\u002F).\n\n> [!TIP]\n> SWE-smith requires Docker to create execution environments. SWE-smith was developed and tested on Ubuntu 22.04.4 LTS.\n> We do *not* plan on supporting Windows or MacOS.\n\nYou can then build a dataset for the repository by...\n1. [Creating an environment](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fenv_construction\u002F#create-an-execution-environment)\n2. [Synthesizing task instances](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fcreate_instances\u002F)\n3. [Keep tasks that break 1+ unit tests](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fharnesses\u002F)\n4. [Generating issue text for your tasks](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fissue_gen\u002F)\n\n## 🏋️ Train SWE-agent's\nTraining SWE-agent's using the [SWE-smith dataset](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FSWE-bench\u002FSWE-smith) is super simple.\n```python\nfrom swesmith.profiles import registry\nfrom datasets import load_dataset\nds = load_dataset(\"SWE-bench\u002FSWE-smith\", split=\"train\") # Loads all 52k task instances\nfor task in ds:\n    rp = registry.get_from_inst(task)  # Get the RepoProfile for the task\n    container = rp.get_container(task) # Returns pointer to a Docker container with the task initialized\n\n    \"\"\"TODO: Train!\"\"\"\n```\n\nSWE-smith has been used to\n* Fine-tune Qwen 2.5 Coder into SWE-agent-LM-32B (A +32% jump on SWE-bench Verified!) using [SWE-agent](https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FSWE-agent) [[Tutorial](https:\u002F\u002Fswesmith.com\u002Fguides\u002Ftrain_swe_agent\u002F)]\n* Perform GRPO style reinforcement learning using [SkyRL](https:\u002F\u002Fgithub.com\u002FNovaSky-AI\u002FSkyRL)\n\n## 💿 Resources\n* [52k Task Instances](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FSWE-bench\u002FSWE-smith)\n* [SWE-agent-LM-32B](https:\u002F\u002Fhuggingface.co\u002FSWE-bench\u002FSWE-agent-LM-32B); **40.2%** pass@1 on [SWE-bench Verified](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FSWE-bench\u002FSWE-bench_Verified)!\n* [26k SWE-agent Trajectories](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FSWE-bench\u002FSWE-smith-trajectories), including the 5k SWE-agent-LM-32B was trained on.\n* [250+ Environments](https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-smith-envs), one Docker image per repo represented in SWE-smith.\n\nAnd there's more coming!\n\n## 💫 Contributions\nWe're actively working on several follow ups!\nCheck out the [Contributing Guide](CONTRIBUTING.md) for more.\n\nContact Person: [John Yang](https:\u002F\u002Fjohn-b-yang.github.io\u002F), [Kilian Lieret](https:\u002F\u002Flieret.net)\n(Email: [johnby@stanford.edu](mailto:johnby@stanford.edu))\n\n## 🪪 License\nMIT. Check `LICENSE` for more information.\n\n## ✍️ Citation\n\n```bibtex\n@inproceedings{yang2025swesmith,\n  title={SWE-smith: Scaling Data for Software Engineering Agents}, \n  author={John Yang and Kilian Lieret and Carlos E. Jimenez and Alexander Wettig and Kabir Khandpur and Yanzhe Zhang and Binyuan Hui and Ofir Press and Ludwig Schmidt and Diyi Yang},\n  booktitle = {Proceedings of the 39th Annual Conference on Neural Information Processing Systems (NeurIPS 2025 D&B Spotlight)},\n  year={2025},\n  eprint={2504.21798},\n  archivePrefix={arXiv},\n  primaryClass={cs.SE},\n  url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.21798},\n  note={arXiv:2504.21798, accepted at NeurIPS 2025 (Spotlight)}\n}\n```\n\n## 📕 Our Other Projects\n\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-bench\">\u003Cimg src=\"docs\u002Fassets\u002Fswebench_logo_text_below.svg\" alt=\"SWE-bench\" height=\"120px\">\u003C\u002Fa>\n    \n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FSWE-agent\">\u003Cimg src=\"docs\u002Fassets\u002Fsweagent_logo_text_below.svg\" alt=\"SWE-agent\" height=\"120px\">\u003C\u002Fa>\n    \n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FMini-SWE-Agent\">\u003Cimg src=\"docs\u002Fassets\u002Fmini_logo_text_below.svg\" alt=\"Mini-SWE-Agent\" height=\"120px\">\u003C\u002Fa>\n    \n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FSWE-ReX\">\u003Cimg src=\"docs\u002Fassets\u002Fswerex_logo_text_below.svg\" alt=\"SWE-ReX\" height=\"120px\">\u003C\u002Fa>\n    \n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSWE-bench\u002Fsb-cli\">\u003Cimg src=\"docs\u002Fassets\u002Fsbcli_logo_text_below.svg\" alt=\"sb-cli\" height=\"120px\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n","\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fswesmith.com\u002F\">\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FSWE-bench_SWE-smith_readme_a9684cb455f4.png\" style=\"height: 10em\" alt=\"Kawhi 这位 SWE-smith（软件工程师铁匠）\" \u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\u003Cbr>\n\u003Cdiv align=\"center\">\n\u003Cstrong>NeurIPS 2025 数据集与基准测试轨道 - 焦点展示 🔦\u003C\u002Fstrong>\n\u003C\u002Fdiv>\n\u003Cbr>\n\u003Cdiv align=\"center\">\n\u003Ca href=\"https:\u002F\u002Fwww.python.org\u002F\">\n  \u003Cimg alt=\"构建\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.10+-1f425f.svg?color=purple\">\n\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fcopyright.princeton.edu\u002Fpolicy\">\n  \u003Cimg alt=\"许可证\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-blue\">\n\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fswesmith\">\n  \u003Cimg src=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fswesmith.svg\">\n\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.21798\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FarXiv-2504.21798-b31b1b.svg\">\n\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Chr \u002F>\n\nSWE-smith 是一个用于训练 [SWE-agents（软件工程师智能体）](https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FSWE-agent) 的工具包。您可以：\n* 将任何 GitHub 仓库转化为 [SWE-gym（软件工程环境）](https:\u002F\u002Fgithub.com\u002FSWE-Gym\u002FSWE-Gym)。\n* 为该仓库创建*无限*的任务（例如，文件定位、程序修复、[SWE-bench](https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-bench)）。\n* 训练一个 LM（语言模型）以成为更优秀的 SWE（软件工程师）([SWE-agent-LM-32B](https:\u002F\u002Fhuggingface.co\u002FSWE-bench\u002FSWE-agent-LM-32B))。\n\n## ⚒️ 构建环境\n如果您有兴趣将 GitHub 仓库转化为 SWE-gym，请从 [源代码](https:\u002F\u002Fswesmith.com\u002Fgetting_started\u002Finstallation\u002F) 安装该软件包。\n\n> [!TIP]\n> SWE-smith 需要 Docker 来创建执行环境。SWE-smith 是在 Ubuntu 22.04.4 LTS 上开发和测试的。\n> 我们*不*计划支持 Windows 或 MacOS。\n\n然后您可以通过以下方式构建该仓库的数据集...\n1. [创建环境](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fenv_construction\u002F#create-an-execution-environment)\n2. [合成任务实例](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fcreate_instances\u002F)\n3. [保留破坏 1+ 单元测试的任务](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fharnesses\u002F)\n4. [为您的任务生成问题文本](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fissue_gen\u002F)\n\n## 🏋️ 训练 SWE-agents\n使用 [SWE-smith 数据集](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FSWE-bench\u002FSWE-smith) 训练 SWE-agents 非常简单。\n```python\nfrom swesmith.profiles import registry\nfrom datasets import load_dataset\nds = load_dataset(\"SWE-bench\u002FSWE-smith\", split=\"train\") # Loads all 52k task instances\nfor task in ds:\n    rp = registry.get_from_inst(task)  # Get the RepoProfile for the task\n    container = rp.get_container(task) # Returns pointer to a Docker container with the task initialized\n\n    \"\"\"TODO: Train!\"\"\"\n```\n\nSWE-smith 已被用于\n* 使用 [SWE-agent](https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FSWE-agent) 微调 Qwen 2.5 Coder 至 SWE-agent-LM-32B（在 SWE-bench Verified 上提升 +32%！）[[教程](https:\u002F\u002Fswesmith.com\u002Fguides\u002Ftrain_swe_agent\u002F)]\n* 使用 [SkyRL](https:\u002F\u002Fgithub.com\u002FNovaSky-AI\u002FSkyRL) 进行 GRPO 风格的强化学习\n\n## 💿 资源\n* [52k 任务实例](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FSWE-bench\u002FSWE-smith)\n* [SWE-agent-LM-32B](https:\u002F\u002Fhuggingface.co\u002FSWE-bench\u002FSWE-agent-LM-32B)；在 [SWE-bench Verified](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FSWE-bench\u002FSWE-bench_Verified) 上 **40.2%** pass@1！\n* [26k SWE-agent 轨迹](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FSWE-bench\u002FSWE-smith-trajectories)，包括 SWE-agent-LM-32B 训练的 5k 数据。\n* [250+ 环境](https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-smith-envs)，每个 SWE-smith 中代表的仓库对应一个 Docker 镜像。\n\n还有更多即将推出！\n\n## 💫 贡献\n我们正在积极开发多个后续项目！\n查看 [贡献指南](CONTRIBUTING.md) 了解更多。\n\n联系人：[John Yang](https:\u002F\u002Fjohn-b-yang.github.io\u002F)，[Kilian Lieret](https:\u002F\u002Flieret.net)\n(邮箱：[johnby@stanford.edu](mailto:johnby@stanford.edu))\n\n## 🪪 许可证\nMIT。查看 `LICENSE` 获取更多信息。\n\n## ✍️ 引用\n\n```bibtex\n@inproceedings{yang2025swesmith,\n  title={SWE-smith: Scaling Data for Software Engineering Agents}, \n  author={John Yang and Kilian Lieret and Carlos E. Jimenez and Alexander Wettig and Kabir Khandpur and Yanzhe Zhang and Binyuan Hui and Ofir Press and Ludwig Schmidt and Diyi Yang},\n  booktitle = {Proceedings of the 39th Annual Conference on Neural Information Processing Systems (NeurIPS 2025 D&B Spotlight)},\n  year={2025},\n  eprint={2504.21798},\n  archivePrefix={arXiv},\n  primaryClass={cs.SE},\n  url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.21798},\n  note={arXiv:2504.21798, accepted at NeurIPS 2025 (Spotlight)}\n}\n```\n\n## 📕 我们的其他项目\n\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-bench\">\u003Cimg src=\"docs\u002Fassets\u002Fswebench_logo_text_below.svg\" alt=\"SWE-bench\" height=\"120px\">\u003C\u002Fa>\n    \n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FSWE-agent\">\u003Cimg src=\"docs\u002Fassets\u002Fsweagent_logo_text_below.svg\" alt=\"SWE-agent\" height=\"120px\">\u003C\u002Fa>\n    \n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FMini-SWE-Agent\">\u003Cimg src=\"docs\u002Fassets\u002Fmini_logo_text_below.svg\" alt=\"Mini-SWE-Agent\" height=\"120px\">\u003C\u002Fa>\n    \n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FSWE-ReX\">\u003Cimg src=\"docs\u002Fassets\u002Fswerex_logo_text_below.svg\" alt=\"SWE-ReX\" height=\"120px\">\u003C\u002Fa>\n    \n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSWE-bench\u002Fsb-cli\">\u003Cimg src=\"docs\u002Fassets\u002Fsbcli_logo_text_below.svg\" alt=\"sb-cli\" height=\"120px\">\u003C\u002Fa>\n\u003C\u002Fdiv>","# SWE-smith 快速上手指南\n\nSWE-smith 是一个用于训练 [SWE-agents](https:\u002F\u002Fgithub.com\u002FSWE-agent\u002FSWE-agent) 的工具包。它可以将任何 GitHub 仓库转化为 SWE-gym，创建无限数量的任务（如文件定位、程序修复），并训练语言模型成为更优秀的软件工程师代理。\n\n## 1. 环境准备\n\n在开始之前，请确保您的开发环境满足以下要求：\n\n*   **操作系统**: 推荐使用 **Ubuntu 22.04.4 LTS**。\n    *   *注意：官方明确表示暂不支持 Windows 或 MacOS。*\n*   **Python 版本**: 需要 **Python 3.10+**。\n*   **核心依赖**: 必须安装 **Docker**，因为 SWE-smith 依赖 Docker 来创建执行环境。\n\n## 2. 安装步骤\n\n您可以通过 PyPI 直接安装该工具包，也可以从源码进行安装。\n\n```bash\n# 推荐方式：通过 pip 安装\npip install swesmith\n\n# 或者从源码安装（如需参与开发或获取最新功能）\ngit clone https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-smith.git\ncd SWE-smith\npip install -e .\n```\n\n## 3. 基本使用\n\n### 构建数据集流程\n\n若要将 GitHub 仓库转化为 SWE-gym 并构建数据集，请遵循以下步骤：\n\n1.  [创建环境](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fenv_construction\u002F#create-an-execution-environment)\n2.  [合成任务实例](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fcreate_instances\u002F)\n3.  [保留破坏 1+ 单元测试的任务](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fharnesses\u002F)\n4.  [为任务生成问题文本](https:\u002F\u002Fswesmith.com\u002Fguides\u002Fissue_gen\u002F)\n\n### 训练 SWE-agent 示例\n\n使用 SWE-smith 数据集训练 SWE-agent 非常简单。以下是加载数据集并初始化容器的基础代码示例：\n\n```python\nfrom swesmith.profiles import registry\nfrom datasets import load_dataset\nds = load_dataset(\"SWE-bench\u002FSWE-smith\", split=\"train\") # Loads all 52k task instances\nfor task in ds:\n    rp = registry.get_from_inst(task)  # Get the RepoProfile for the task\n    container = rp.get_container(task) # Returns pointer to a Docker container with the task initialized\n\n    \"\"\"TODO: Train!\"\"\"\n```\n\n### 资源链接\n\n*   **数据集**: [52k Task Instances](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FSWE-bench\u002FSWE-smith)\n*   **预训练模型**: [SWE-agent-LM-32B](https:\u002F\u002Fhuggingface.co\u002FSWE-bench\u002FSWE-agent-LM-32B)\n*   **轨迹数据**: [26k SWE-agent Trajectories](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FSWE-bench\u002FSWE-smith-trajectories)","某金融科技公司的研发团队希望为内部遗留的 Python 核心系统训练一个专属的代码修复 AI 助手，以应对日益增长的维护需求。\n\n### 没有 SWE-smith 时\n- 团队需人工梳理 Git 提交记录来构造训练数据，耗时数月且覆盖率严重不足。\n- 缺乏统一的执行环境，不同模块依赖冲突导致无法自动化验证修复结果的有效性。\n- 通用开源模型不懂内部业务逻辑，生成的代码常出现幻觉或引入新的安全漏洞。\n- 每次尝试新算法都要重新配置复杂的 Docker 镜像，环境搭建成为研发效率的最大瓶颈。\n\n### 使用 SWE-smith 后\n- SWE-smith 自动将仓库转化为 SWE-gym，快速生成包含 52k 任务的标准化高质量数据集。\n- 内置的 Harness 机制自动运行单元测试，精准过滤无效任务并保证每个样本的可解性。\n- 利用合成数据微调模型，使 Agent 在私有代码库上的 Pass@1 通过率显著提升 32%。\n- 预置的 Docker 环境模板消除了依赖配置难题，让实验迭代周期从数周缩短至数天。\n\nSWE-smith 通过规模化数据工程能力，让企业能够低成本构建高性能专属代码智能体，彻底改变了传统 AI 辅助开发的训练模式。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FSWE-bench_SWE-smith_a9684cb4.png","SWE-bench","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FSWE-bench_8c8506e2.png","Organization for maintaining SWE-bench and related projects",null,"https:\u002F\u002Fswebench.com\u002F","https:\u002F\u002Fgithub.com\u002FSWE-bench",[82,86,90,94,98,102,105,109,112,116],{"name":83,"color":84,"percentage":85},"Python","#3572A5",94.1,{"name":87,"color":88,"percentage":89},"Go","#00ADD8",1.7,{"name":91,"color":92,"percentage":93},"C#","#178600",1.4,{"name":95,"color":96,"percentage":97},"C","#555555",0.9,{"name":99,"color":100,"percentage":101},"Shell","#89e051",0.4,{"name":103,"color":104,"percentage":101},"Ruby","#701516",{"name":106,"color":107,"percentage":108},"C++","#f34b7d",0.3,{"name":110,"color":111,"percentage":108},"Rust","#dea584",{"name":113,"color":114,"percentage":115},"Java","#b07219",0.2,{"name":117,"color":118,"percentage":119},"PHP","#4F5D95",0.1,615,114,"2026-04-04T18:09:00","MIT","Linux","未说明",{"notes":127,"python":128,"dependencies":129},"必须安装 Docker 以创建执行环境；仅在 Ubuntu 22.04.4 LTS 上开发和测试过，明确不支持 Windows 和 macOS；主要用于生成任务实例及训练 SWE-agent 大语言模型。","3.10+",[130],"datasets",[13,15,26],[133,134,135,136],"agents","language-model","software-engineering","training","2026-03-27T02:49:30.150509","2026-04-06T07:12:53.015126",[140,145,149,154,158,163],{"id":141,"question_zh":142,"answer_zh":143,"source_url":144},3626,"最近的评估变更导致 Docker 镜像中的 commit 与实例 ID 不匹配怎么办？","该问题已在 #85 中解决。注意 HuggingFace 上的 SWE-smith 数据集可能缺少 \"KEY_PATCH_TEST\" 字段，这会影响新的评估方法。Docker 镜像已更新，实例 ID 对应的 commit 现在是不含 F2P 测试用例的版本。","https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-smith\u002Fissues\u002F110",{"id":146,"question_zh":147,"answer_zh":148,"source_url":144},3627,"如何在数据集中获取原始的 Bug Patch 提交？","数据集可能不直接包含原始补丁。维护者表示可以推送一个包含每个轨迹补丁的版本。如果需要实际创建的补丁，请关注后续更新或通过 `--f2p_only` 标志了解最新功能。",{"id":150,"question_zh":151,"answer_zh":152,"source_url":153},3628,"如何运行 SWE-Smith 的健全性检查（sanity check）？","可以使用以下命令：`python swesmith\u002Fharness\u002Feval.py --instance_ids \u003C...> --run_id check`。也可以使用发布的 Docker 镜像运行对应测试，例如 `docker run -it --rm jyangballin\u002Fswesmith.x86_64.spulec_1776_freezegun.5f171db0`。","https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-smith\u002Fissues\u002F26",{"id":155,"question_zh":156,"answer_zh":157,"source_url":153},3629,"遇到测试不稳定（flaky tests）或金标准补丁无法通过测试如何处理？","如果存在不稳定的任务实例，请在新 issue 中指出，维护者会通过 `eval.py` 命令进行健全性检查。部分已知错误（如 `hydra` 插件未注册）可能会被忽略。",{"id":159,"question_zh":160,"answer_zh":161,"source_url":162},3630,"在 AWS EC2 上运行 SWE-agent 时出现 Docker 构建错误如何解决？","请尝试升级依赖包：`pip install --upgrade swe-rex mini-swe-agent`。更新 `swe-rex` 到最新版本（如 1.4.0）通常能解决此问题，相关问题可能源于 Python Docker 镜像的更新。","https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-smith\u002Fissues\u002F140",{"id":164,"question_zh":165,"answer_zh":166,"source_url":167},3631,"哪里可以下载实验轨迹以复现评估结果？","轨迹可在 https:\u002F\u002Fgithub.com\u002Fswe-bench\u002Fexperiments\u002F 下载。建议使用 `sweagent i` 工具对比轨迹文件来调试性能差异，查看具体动作以排查环境配置问题。","https:\u002F\u002Fgithub.com\u002FSWE-bench\u002FSWE-smith\u002Fissues\u002F100",[169,172,175,178],{"id":170,"version":171,"summary_zh":78,"released_at":78},103221,"v0.0.6",{"id":173,"version":174,"summary_zh":78,"released_at":78},103222,"v0.0.5",{"id":176,"version":177,"summary_zh":78,"released_at":78},103223,"v0.0.2",{"id":179,"version":180,"summary_zh":78,"released_at":78},103224,"v0.0.1"]