[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tool-EnzymeAD--Enzyme.jl":3,"similar-EnzymeAD--Enzyme.jl":186},{"id":4,"github_repo":5,"name":6,"description_en":7,"description_zh":8,"ai_summary_zh":8,"readme_en":9,"readme_zh":10,"quickstart_zh":11,"use_case_zh":12,"hero_image_url":13,"owner_login":14,"owner_name":15,"owner_avatar_url":16,"owner_bio":17,"owner_company":18,"owner_location":18,"owner_email":19,"owner_twitter":18,"owner_website":20,"owner_url":21,"languages":22,"stars":27,"forks":28,"last_commit_at":29,"license":30,"difficulty_score":31,"env_os":32,"env_gpu":33,"env_ram":32,"env_deps":34,"category_tags":39,"github_topics":42,"view_count":51,"oss_zip_url":18,"oss_zip_packed_at":18,"status":52,"created_at":53,"updated_at":54,"faqs":55,"releases":85},2180,"EnzymeAD\u002FEnzyme.jl","Enzyme.jl","Julia bindings for the Enzyme automatic differentiator","Enzyme.jl 是 Julia 语言的高性能自动微分工具，作为底层引擎 Enzyme 的官方接口，它能让计算机自动计算复杂函数的导数。在机器学习、科学计算和优化算法中，手动推导梯度不仅繁琐且容易出错，而传统自动微分工具往往难以兼顾速度与灵活性。Enzyme.jl 正是为了解决这一痛点而生，它直接对编译后的 LLVM 中间代码进行分析，无需修改源代码即可生成高效的梯度计算逻辑。\n\n这款工具特别适合科研人员、算法工程师以及高性能计算开发者使用，尤其是那些需要在 Julia 生态中构建深度学习模型或求解微分方程的用户。其核心亮点在于“极致性能”：由于能直接对经过编译器优化后的代码进行微分，Enzyme.jl 的运行效率通常优于现有的主流自动微分方案，甚至能无缝支持 GPU 加速和复杂的控制流。通过简单的 `autodiff` 调用，用户即可轻松获得精确的梯度结果，将精力从繁琐的数学推导中解放出来，专注于核心算法的创新与设计。","# \u003Cimg src=\"https:\u002F\u002Fenzyme.mit.edu\u002Flogo.svg\" width=\"75\" align=left> The Enzyme High-Performance Automatic Differentiator of LLVM\n\n[![Stable](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-stable-blue.svg)](https:\u002F\u002Fenzyme.mit.edu\u002Fjulia\u002Fstable)\n[![Dev](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-dev-blue.svg)](https:\u002F\u002Fenzyme.mit.edu\u002Fjulia\u002Fdev)\n[![Build Status](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fworkflows\u002FCI\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Factions)\n[![Coverage](https:\u002F\u002Fcodecov.io\u002Fgh\u002FEnzymeAD\u002FEnzyme.jl\u002Fbranch\u002Fmain\u002Fgraph\u002Fbadge.svg)](https:\u002F\u002Fcodecov.io\u002Fgh\u002FEnzymeAD\u002FEnzyme.jl)\n\nThis is a package containing the Julia bindings for [Enzyme](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002Fenzyme). This is very much a work in progress and bug reports\u002Fdiscussion is greatly appreciated!\n\nEnzyme is a plugin that performs automatic differentiation (AD) of statically analyzable LLVM. It is highly-efficient and its ability perform AD on optimized code allows Enzyme to meet or exceed the performance of state-of-the-art AD tools.\n\nEnzyme.jl can be installed in the usual way Julia packages are installed\n```\n] add Enzyme\n```\n\nEnzyme.jl can be used by calling `autodiff` on a function to be differentiated as shown below:\n\n```julia\nusing Enzyme, Test\n\nf1(x) = x*x\n# Returns a tuple of active returns, which in this case is simply (2.0,)\n@test first(autodiff(Reverse, f1, Active(1.0))[1]) ≈ 2.0\n```\n\nFor details, see the [package documentation](https:\u002F\u002Fenzyme.mit.edu\u002Fjulia).\n\nMore information on installing and using Enzyme directly (not through Julia) can be found on our website: [https:\u002F\u002Fenzyme.mit.edu](https:\u002F\u002Fenzyme.mit.edu).\n\nTo get involved or if you have questions, please join our [mailing list](https:\u002F\u002Fgroups.google.com\u002Fd\u002Fforum\u002Fenzyme-dev).\n\nIf using this code in an academic setting, please cite the following two papers (first for Enzyme as a whole, then for GPU+optimizations):\n```bibtex\n@inproceedings{NEURIPS2020_9332c513,\n author = {Moses, William and Churavy, Valentin},\n booktitle = {Advances in Neural Information Processing Systems},\n editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},\n pages = {12472--12485},\n publisher = {Curran Associates, Inc.},\n title = {Instead of Rewriting Foreign Code for Machine Learning, Automatically Synthesize Fast Gradients},\n url = {https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper\u002F2020\u002Ffile\u002F9332c513ef44b682e9347822c2e457ac-Paper.pdf},\n volume = {33},\n year = {2020}\n}\n@inproceedings{10.1145\u002F3458817.3476165,\n author = {Moses, William S. and Churavy, Valentin and Paehler, Ludger and H\\\"{u}ckelheim, Jan and Narayanan, Sri Hari Krishna and Schanen, Michel and Doerfert, Johannes},\n title = {Reverse-Mode Automatic Differentiation and Optimization of GPU Kernels via Enzyme},\n year = {2021},\n isbn = {9781450384421},\n publisher = {Association for Computing Machinery},\n address = {New York, NY, USA},\n url = {https:\u002F\u002Fdoi.org\u002F10.1145\u002F3458817.3476165},\n doi = {10.1145\u002F3458817.3476165},\n booktitle = {Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis},\n articleno = {61},\n numpages = {16},\n keywords = {CUDA, LLVM, ROCm, HPC, AD, GPU, automatic differentiation},\n location = {St. Louis, Missouri},\n series = {SC '21}\n}\n```\n","# \u003Cimg src=\"https:\u002F\u002Fenzyme.mit.edu\u002Flogo.svg\" width=\"75\" align=left> LLVM 的高性能自动微分工具 Enzyme\n\n[![稳定版](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-stable-blue.svg)](https:\u002F\u002Fenzyme.mit.edu\u002Fjulia\u002Fstable)\n[![开发版](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-dev-blue.svg)](https:\u002F\u002Fenzyme.mit.edu\u002Fjulia\u002Fdev)\n[![构建状态](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fworkflows\u002FCI\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Factions)\n[![覆盖率](https:\u002F\u002Fcodecov.io\u002Fgh\u002FEnzymeAD\u002FEnzyme.jl\u002Fbranch\u002Fmain\u002Fgraph\u002Fbadge.svg)](https:\u002F\u002Fcodecov.io\u002Fgh\u002FEnzymeAD\u002FEnzyme.jl)\n\n这是一个包含 [Enzyme](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002Fenzyme) Julia 绑定的软件包。目前仍处于开发阶段，欢迎提交 bug 报告和参与讨论！\n\nEnzyme 是一个插件，用于对可静态分析的 LLVM 代码进行自动微分（AD）。它效率极高，并且能够在优化后的代码上执行 AD，因此其性能可以达到甚至超越当前最先进的自动微分工具。\n\nEnzyme.jl 可以通过常规的 Julia 包管理方式安装：\n```\n] add Enzyme\n```\n\n使用 Enzyme.jl 时，只需对需要求导的函数调用 `autodiff` 即可，示例如下：\n\n```julia\nusing Enzyme, Test\n\nf1(x) = x*x\n# 返回一个包含活跃返回值的元组，在本例中为 (2.0,)\n@test first(autodiff(Reverse, f1, Active(1.0))[1]) ≈ 2.0\n```\n\n更多详细信息请参阅 [包文档](https:\u002F\u002Fenzyme.mit.edu\u002Fjulia)。\n\n有关直接安装和使用 Enzyme（而非通过 Julia）的更多信息，请访问我们的官网：[https:\u002F\u002Fenzyme.mit.edu](https:\u002F\u002Fenzyme.mit.edu)。\n\n如需参与或有任何问题，请加入我们的 [邮件列表](https:\u002F\u002Fgroups.google.com\u002Fd\u002Fforum\u002Fenzyme-dev)。\n\n如果您在学术研究中使用此代码，请引用以下两篇论文（第一篇为 Enzyme 整体介绍，第二篇则聚焦于 GPU 加速及优化技术）：\n```bibtex\n@inproceedings{NEURIPS2020_9332c513,\n author = {Moses, William and Churavy, Valentin},\n booktitle = {Advances in Neural Information Processing Systems},\n editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},\n pages = {12472--12485},\n publisher = {Curran Associates, Inc.},\n title = {Instead of Rewriting Foreign Code for Machine Learning, Automatically Synthesize Fast Gradients},\n url = {https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper\u002F2020\u002Ffile\u002F9332c513ef44b682e9347822c2e457ac-Paper.pdf},\n volume = {33},\n year = {2020}\n}\n@inproceedings{10.1145\u002F3458817.3476165,\n author = {Moses, William S. and Churavy, Valentin and Paehler, Ludger and H\\\"{u}ckelheim, Jan and Narayanan, Sri Hari Krishna and Schanen, Michel and Doerfert, Johannes},\n title = {Reverse-Mode Automatic Differentiation and Optimization of GPU Kernels via Enzyme},\n year = {2021},\n isbn = {9781450384421},\n publisher = {Association for Computing Machinery},\n address = {New York, NY, USA},\n url = {https:\u002F\u002Fdoi.org\u002F10.1145\u002F3458817.3476165},\n doi = {10.1145\u002F3458817.3476165},\n booktitle = {Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis},\n articleno = {61},\n numpages = {16},\n keywords = {CUDA, LLVM, ROCm, HPC, AD, GPU, automatic differentiation},\n location = {St. Louis, Missouri},\n series = {SC '21}\n}\n```","# Enzyme.jl 快速上手指南\n\nEnzyme.jl 是高性能自动微分工具 Enzyme 的 Julia 语言绑定。它基于 LLVM 进行静态分析，能够对优化后的代码执行高效的自动微分（AD），性能可媲美甚至超越当前最先进的 AD 工具。\n\n## 环境准备\n\n*   **操作系统**：支持 Linux、macOS 和 Windows。\n*   **前置依赖**：\n    *   已安装 **Julia** (推荐最新稳定版)。\n    *   无需额外安装 LLVM 或 C++ 编译器，Enzyme.jl 会自动处理相关二进制依赖。\n*   **网络建议**：由于 Julia 包服务器位于海外，国内用户建议在安装前配置国内镜像源以加速下载。\n\n    ```julia\n    using Pkg\n    Pkg.Registry.add(\"General\")\n    # 可选：配置清华或中科大镜像加速包下载\n    ENV[\"JULIA_PKG_SERVER\"] = \"https:\u002F\u002Fmirrors.tuna.tsinghua.edu.cn\u002Fjulia\"\n    ```\n\n## 安装步骤\n\n在 Julia 的 REPL（交互式命令行）中，进入包管理模式并执行以下命令：\n\n```julia\n] add Enzyme\n```\n\n安装完成后，即可在项目中通过 `using Enzyme` 加载使用。\n\n## 基本使用\n\nEnzyme.jl 的核心用法是调用 `autodiff` 函数对目标函数进行微分。以下是一个计算 $f(x) = x^2$ 导数的最简示例：\n\n```julia\nusing Enzyme, Test\n\n# 定义待微分的函数\nf1(x) = x*x\n\n# 执行反向模式自动微分\n# Active(1.0) 表示输入值 1.0 是需要计算梯度的活跃变量\n# 返回值是一个元组，包含活跃变量的梯度结果\nresult = autodiff(Reverse, f1, Active(1.0))\n\n# 提取并验证结果 (预期结果为 2.0)\n@test first(result[1]) ≈ 2.0\n\nprintln(\"导数结果：\", first(result[1]))\n```\n\n**说明：**\n*   `Reverse`：指定使用反向模式微分（适用于标量输出、多输入场景）。\n*   `Active`：标记需要计算梯度的参数。\n*   结果通常以元组形式返回，需根据具体结构提取梯度值。\n\n更多高级用法（如混合模式、GPU 支持、自定义规则）请参阅 [官方文档](https:\u002F\u002Fenzyme.mit.edu\u002Fjulia)。","一位计算物理学家正在使用 Julia 开发一套高精度的流体动力学仿真程序，需要频繁对复杂的数值积分函数进行梯度计算以优化参数。\n\n### 没有 Enzyme.jl 时\n- **手动推导易出错**：面对包含数千行逻辑的复杂物理方程，人工推导导数公式极易出现符号错误，且调试成本极高。\n- **性能瓶颈明显**：使用传统的基于操作符重载的自动微分库（如 ForwardDiff）会引入大量临时内存分配，导致在大规模网格计算时运行速度显著下降。\n- **代码重构困难**：为了适配某些特定的微分工具，不得不将原本高度优化的 LLVM 底层代码或 GPU 内核重写为受限的子集，破坏了原有架构。\n- **迭代周期漫长**：每次修改物理模型后，都需要重新验证导数正确性并调整性能，严重拖慢了科研实验的迭代进度。\n\n### 使用 Enzyme.jl 后\n- **零样板代码生成**：只需对现有的高性能 Julia 函数调用 `autodiff`，Enzyme.jl 即可直接在 LLVM 中间码层面合成精确的梯度代码，无需手动推导。\n- **极致运行效率**：得益于对优化后代码的直接微分能力，生成的梯度函数几乎无额外开销，在基准测试中性能媲美甚至超越手写优化的 C++\u002FCUDA 代码。\n- **无缝兼容现有架构**：完美支持复杂的控制流、递归调用及 GPU 内核，开发者无需为了微分而牺牲代码结构或放弃底层优化特性。\n- **加速科研创新**：将梯度计算的开发时间从数天缩短至几分钟，让研究者能专注于物理模型本身的改进而非数学实现的细节。\n\nEnzyme.jl 通过在编译器层面实现高效自动微分，让科学家在不牺牲性能的前提下，轻松获得复杂科学计算程序的精确梯度。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FEnzymeAD_Enzyme.jl_51aa0135.png","EnzymeAD","Enzyme Automatic Differentiation Compiler","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FEnzymeAD_519ef65a.png","",null,"enzyme-dev@googlegroups.com","https:\u002F\u002Fenzyme.mit.edu","https:\u002F\u002Fgithub.com\u002FEnzymeAD",[23],{"name":24,"color":25,"percentage":26},"Julia","#a270ba",100,556,91,"2026-04-02T03:32:58","MIT",1,"未说明","非必需。支持通过 LLVM 后端对 GPU 内核（如 CUDA、ROCm）进行自动微分，具体取决于底层 LLVM 配置",{"notes":35,"python":36,"dependencies":37},"该工具是 Enzyme (LLVM 自动微分插件) 的 Julia 绑定包。需安装 Julia 并通过包管理器 (`] add Enzyme`) 安装。不支持直接通过 Python 调用。若需进行 GPU 微分，宿主环境需具备相应的 GPU 编译工具链（如 CUDA 或 ROCm）。","不适用 (基于 Julia)",[24,38],"LLVM",[40,41],"开发框架","插件",[43,44,45,46,47,48,49,50],"enzyme","julia","ad","machine-learning","llvm","compiler","automatic-differentiation","differentiable-programming",2,"ready","2026-03-27T02:49:30.150509","2026-04-06T06:53:22.726697",[56,61,66,71,76,81],{"id":57,"question_zh":58,"answer_zh":59,"source_url":60},10049,"在 Turing 模型中使用 Enzyme 时遇到段错误（segfault）或警告怎么办？","这通常是由于 Enzyme 版本过旧导致的。请尝试更新 Enzyme 到最新补丁版本。维护者指出，相关修复已合并（如 PR #934），更新后示例代码应能正常运行。如果问题依旧，可以尝试禁用类型警告并启用运行时活动检查作为临时方案：\n```julia\nEnzyme.API.runtimeActivity!(true)\nEnzyme.API.typeWarning!(false)\n```\n此外，确保在 Linux 环境下测试，因为某些特定于平台的错误可能在其他系统上表现不同。","https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fissues\u002F650",{"id":62,"question_zh":63,"answer_zh":64,"source_url":65},10050,"为什么在 Julia 1.11 中使用矩阵输入计算 HVP（Hessian-Vector Product）时结果不正确（只有第一行有值）？","这是一个已知问题，但在后续版本中已被修复。如果您遇到此问题，请确保升级到最新的 Enzyme 版本。维护者确认该问题已在某个时间点解决。如果升级后问题仍然存在，请检查您的 LLVM 模块配置，确保 `llvm_module2` 是字面量字符串，并且移除了不必要的 `@` 符号。","https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fissues\u002F2071",{"id":67,"question_zh":68,"answer_zh":69,"source_url":70},10051,"在使用 SymbolicRegression 或其他包配合 Enzyme 时遇到\"nested task error: task switch not allowed\"错误如何解决？","该错误通常由旧版本的 Enzyme 引起，相关修复已在最近的发布中标记并推出。请首先确认您使用的 Enzyme 版本是否为最新版。维护者建议升级 Enzyme 以解决此任务切换错误。如果升级后问题消失，说明是版本兼容性问题；若需防止回归，可以将包含大型选项结构体（如 `Options()`）的场景添加到集成测试中。","https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fissues\u002F2081",{"id":72,"question_zh":73,"answer_zh":74,"source_url":75},10052,"对 Gridap 等复杂包代码进行微分时出现\"Mismatched activity\"错误或段错误怎么办？","当看到类似\"Mismatched activity for: store...\"的错误信息时，通常意味着您将一个常量变量用作活跃内存的临时存储。解决方案有两种：\n1. 重构代码，避免该变量成为条件活跃变量。\n2. 作为临时变通方法，启用运行时活动检查：\n```julia\nEnzyme.API.runtimeActivity!(true)\n```\n如果应用上述方法后仍无法解决问题且无法复现，可能是特定环境下的偶发问题，建议提供最小可复现示例以便进一步排查。","https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fissues\u002F447",{"id":77,"question_zh":78,"answer_zh":79,"source_url":80},10053,"如何正确配置 TagBot 以自动触发注册表更新？","您需要更新项目根目录下的 `.github\u002Fworkflows\u002FTagBot.yml` 文件，包含 issue comment 触发器。具体配置指令和详细信息请参考 Julia Discourse 论坛的相关公告帖子：https:\u002F\u002Fdiscourse.julialang.org\u002Ft\u002Fann-required-updates-to-tagbot-yml\u002F49249。配置完成后，可以通过在 Issue 中评论来手动触发 TagBot。","https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fissues\u002F24",{"id":82,"question_zh":83,"answer_zh":84,"source_url":60},10054,"遇到\"UndefVarError\"且堆栈跟踪指向 Enzyme.Compiler 时该如何处理？","此类错误有时与特定的代码模式或变量作用域有关。在某些案例中（如 Turing 采样），即使禁用了警告，也可能因内部编译器状态导致未定义变量错误。建议首先尝试更新 Enzyme 到最新版本，因为许多编译器相关的 bug 会在新版本中修复。如果问题持续，尝试简化模型或使用 `Enzyme.API.runtimeActivity!(true)` 来改变编译行为，看是否能绕过该错误。",[86,91,96,101,106,111,116,121,126,131,136,141,146,151,156,161,166,171,176,181],{"id":87,"version":88,"summary_zh":89,"released_at":90},107283,"v0.13.138","## Enzyme v0.13.138\n\n[Diff since v0.13.137](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.137...v0.13.138)\n\n\nThis release has been identified as a backport.\nAutomated changelogs for backports tend to be wildly incorrect.\nTherefore, the list of issues and pull requests is hidden.\n\u003C!--\n\n\n-->","2026-04-02T05:41:07",{"id":92,"version":93,"summary_zh":94,"released_at":95},107284,"v0.13.137","## Enzyme v0.13.137\n\n[Diff since v0.13.136](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.136...v0.13.137)\n\n\nThis release has been identified as a backport.\nAutomated changelogs for backports tend to be wildly incorrect.\nTherefore, the list of issues and pull requests is hidden.\n\u003C!--\n\n**Closed issues:**\n- MixedReturnException in runtime_generic_augfwd with custom rule on function returning complex type (#3021)\n\n-->","2026-03-31T23:02:16",{"id":97,"version":98,"summary_zh":99,"released_at":100},107285,"v0.13.136","## Enzyme v0.13.136\n\n[Diff since v0.13.135](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.135...v0.13.136)\n\n\n**Merged pull requests:**\n- Remove closure (#3029) (@wsmoses)\n- Fix condition where a byval argument convention means no additional r… (#3030) (@wsmoses)\n\n**Closed issues:**\n- Reverse AD through a function with sum on CUDA backend leads to scary crashes (#3025)","2026-03-29T07:24:32",{"id":102,"version":103,"summary_zh":104,"released_at":105},107286,"v0.13.135","## Enzyme v0.13.135\n\n[Diff since v0.13.134](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.134...v0.13.135)\n\n\n**Merged pull requests:**\n- Cleanup, reduce memory\u002Fgc, and maybe speedup compilation (#3023) (@wsmoses)\n- Fix deepcopy on 1.10 (#3027) (@wsmoses)\n\n**Closed issues:**\n- Testing rule errors out on 1.10 but works fine on 1.12 (#2905)\n- two errors on julia 1.12 for Flux layers (#2998)\n- please advise: where to cache intermediate result (#3017)\n- Reverse differentiation through function map simulation with threading (#3022)","2026-03-28T17:28:18",{"id":107,"version":108,"summary_zh":109,"released_at":110},107287,"v0.13.134","## Enzyme v0.13.134\n\n[Diff since v0.13.133](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.133...v0.13.134)\n\n\n**Merged pull requests:**\n- Fixup new callingconv to be newpm (#3011) (@wsmoses)\n- Syrkbc (#3015) (@wsmoses)\n- More deepcopy tests (#3018) (@wsmoses)\n\n**Closed issues:**\n- Reverse-mode generates DSYMM with invalid LDB for Adjoint * Matrix, gradient 2x too small (#2999)\n- Compiler(?) error on 1.10 (#3003)\n- StackOverflowError differentiating NonlinearProblem with custom struct parameter (#3013)","2026-03-23T16:33:33",{"id":112,"version":113,"summary_zh":114,"released_at":115},107288,"v0.13.133","## Enzyme v0.13.133\n\n[Diff since v0.13.132](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.132...v0.13.133)\n\n\n**Merged pull requests:**\n- Bump version to 0.13.133 and update Enzyme_jll (#3007) (@wsmoses)","2026-03-22T15:11:04",{"id":117,"version":118,"summary_zh":119,"released_at":120},107289,"v0.13.132","## Enzyme v0.13.132\n\n[Diff since v0.13.131](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.131...v0.13.132)","2026-03-18T19:08:27",{"id":122,"version":123,"summary_zh":124,"released_at":125},107290,"v0.13.131","## Enzyme v0.13.131\n\n[Diff since v0.13.130](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.130...v0.13.131)\n\n\n**Merged pull requests:**\n- Fix mixed fn jit (#2996) (@wsmoses)\n\n**Closed issues:**\n- `Fix1` throws MethodError on `Duplicated(::Base.Fix1{...}, ::Base.RefValue{Base.Fix1{...}})` (or worse) (#2994)","2026-03-12T20:19:15",{"id":127,"version":128,"summary_zh":129,"released_at":130},107291,"v0.13.130","## Enzyme v0.13.130\n\n[Diff since v0.13.129](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.129...v0.13.130)\n\n\n**Merged pull requests:**\n- Add easy_rule for BigFloat division (#2934) (@kshyatt)\n- Add SymbolicRegression integration test (#2965) (@MilesCranmerBot)\n- Add inactive rules for timing, Printf, GC, and unicode case functions (#2972) (@michel2323)\n- Add bibtex marker to README.md (#2975) (@abhro)\n- Ignore all Manifest.toml files in docs\u002F from tracking (#2976) (@abhro)\n- Improve handling of jlroots (#2978) (@wsmoses)\n- Enable Molly integration testing on Julia 1.10 (#2982) (@jgreener64)\n- Bump actions\u002Fupload-artifact from 6 to 7 (#2988) (@dependabot[bot])\n- Bump julia-actions\u002Fcache from 2 to 3 (#2993) (@dependabot[bot])\n- Update Enzyme_jll version to 0.0.251 (#2995) (@wsmoses)\n\n**Closed issues:**\n- Julia v1.11 Linux: LoadError when Enzyme tries to load macOS Accelerate framework (#2502)\n- Julia 1.12 Support Roadmap (#2699)\n- Julia 1.12 error with array views (#2963)\n- LLVM error on 1.12 with Float32 and infinity-norm (#2985)\n- Duplicated() does not accept mixed storage types (#2989)","2026-03-12T16:55:47",{"id":132,"version":133,"summary_zh":134,"released_at":135},107292,"v0.13.129","## Enzyme v0.13.129\n\n[Diff since v0.13.128](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.128...v0.13.129)\n\n\n**Merged pull requests:**\n- Fix substring size handling (#2959) (@wsmoses)","2026-02-01T13:01:39",{"id":137,"version":138,"summary_zh":139,"released_at":140},107293,"v0.13.128","## Enzyme v0.13.128\n\n[Diff since v0.13.127](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.127...v0.13.128)\n\n\n**Merged pull requests:**\n- 1.12: fix blasdot test (#2953) (@wsmoses)\n- Fix enzyme notype (#2954) (@wsmoses)\n- Official 1.12 support (#2957) (@wsmoses)\n- Fix version bound on type analysis error (#2958) (@penelopeysm)\n\n**Closed issues:**\n- Adapt to new root calling convention (#2707)\n- Odd `IllegalTypeAnalysisException` on 1.12 (#2900)\n- 1.12 sparsearrays error (#2946)\n- Bijectors InvertibleBatchNorm 1.12 error (#2955)","2026-02-01T06:44:28",{"id":142,"version":143,"summary_zh":144,"released_at":145},107294,"v0.13.127","## Enzyme v0.13.127\n\n[Diff since v0.13.126](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.126...v0.13.127)\n\n\n**Merged pull requests:**\n- Fix restore alloca during union (#2951) (@wsmoses)","2026-01-31T21:42:58",{"id":147,"version":148,"summary_zh":149,"released_at":150},107295,"v0.13.126","## Enzyme v0.13.126\n\n[Diff since v0.13.125](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.125...v0.13.126)\n\n\n**Merged pull requests:**\n- More struct ordering (#2943) (@wsmoses)\n- Fix abstract rooted type (#2945) (@wsmoses)\n- Mark eqtable put as nofree (#2949) (@wsmoses)\n- CUDA: fix cuptr and curef (#2950) (@wsmoses)\n\n**Closed issues:**\n- Julia 1.12: `IllegalTypeAnalysisException` in MPI integration test (#2933)\n- Forward mode IllegalTypeAnalysisException: @concrete struct with nested struct + DiffCache (#2942)\n- Forward mode MethodError: Cannot convert Type{NamedTuple} to DataType in equivalent_rooted_type (#2944)","2026-01-31T12:10:03",{"id":152,"version":153,"summary_zh":154,"released_at":155},107296,"v0.13.125","## Enzyme v0.13.125\n\n[Diff since v0.13.124](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.124...v0.13.125)\n\n\n\n**Closed issues:**\n- Forward mode IllegalTypeAnalysisException with nested structs (#2940)","2026-01-29T21:06:11",{"id":157,"version":158,"summary_zh":159,"released_at":160},107297,"v0.13.124","## Enzyme v0.13.124\n\n[Diff since v0.13.123](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.123...v0.13.124)\n\n\n**Merged pull requests:**\n- Fix ordering nested structs (#2941) (@wsmoses)","2026-01-29T06:01:45",{"id":162,"version":163,"summary_zh":164,"released_at":165},107298,"v0.13.123","## Enzyme v0.13.123\n\n[Diff since v0.13.122](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.122...v0.13.123)\n\n\n**Merged pull requests:**\n- 1.12 invalidation fix (#2939) (@wsmoses)\n\n**Closed issues:**\n- BatchDuplicated Forward mode crashes with closure capturing Vector{Vector{Float64}} (#2936)","2026-01-28T07:44:01",{"id":167,"version":168,"summary_zh":169,"released_at":170},107299,"v0.13.122","## Enzyme v0.13.122\n\n[Diff since v0.13.121](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.121...v0.13.122)\n\n\n**Merged pull requests:**\n- Update Project.toml (#2937) (@wsmoses)\n- fix 1.12 const redefinition in Turing integration test (#2938) (@penelopeysm)","2026-01-27T22:02:35",{"id":172,"version":173,"summary_zh":174,"released_at":175},107300,"v0.13.121","## Enzyme v0.13.121\n\n[Diff since v0.13.120](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.120...v0.13.121)\n\n\n**Merged pull requests:**\n- 1.12: more rule mismatch cconv (#2930) (@wsmoses)\n- Bump version to 0.13.121 and update Enzyme_jll (#2931) (@wsmoses)\n\n**Closed issues:**\n- 1.12 assertion error (#2924)","2026-01-27T06:22:30",{"id":177,"version":178,"summary_zh":179,"released_at":180},107301,"v0.13.120","## Enzyme v0.13.120\n\n[Diff since v0.13.119](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.119...v0.13.120)\n\n\n**Merged pull requests:**\n- 1.12: Compute types before ival (#2928) (@wsmoses)","2026-01-26T19:54:23",{"id":182,"version":183,"summary_zh":184,"released_at":185},107302,"v0.13.119","## Enzyme v0.13.119\n\n[Diff since v0.13.118](https:\u002F\u002Fgithub.com\u002FEnzymeAD\u002FEnzyme.jl\u002Fcompare\u002Fv0.13.118...v0.13.119)\n\n\n**Merged pull requests:**\n- Fix inactive on returnroots for 1.12+ (#2915) (@wsmoses)\n- Add some easy rules for BigFloat (#2916) (@kshyatt)\n- fix broadcast recursion (#2918) (@oscardssmith)\n- CompatHelper: bump compat for Enzyme_jll to 0.0.239, (keep existing compat) (#2920) (@github-actions[bot])\n- Fix strip tracked pointers (#2923) (@wsmoses)\n- 1.12 continued rule callconv (#2926) (@wsmoses)\n- Improve restore alloca (#2927) (@wsmoses)\n\n**Closed issues:**\n- No rule for `mpfr_add` (#2889)","2026-01-26T06:45:12",[187,198,207,215,223,235],{"id":188,"name":189,"github_repo":190,"description_zh":191,"stars":192,"difficulty_score":193,"last_commit_at":194,"category_tags":195,"status":52},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",[40,196,197],"图像","Agent",{"id":199,"name":200,"github_repo":201,"description_zh":202,"stars":203,"difficulty_score":51,"last_commit_at":204,"category_tags":205,"status":52},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,"2026-04-05T11:33:21",[40,197,206],"语言模型",{"id":208,"name":209,"github_repo":210,"description_zh":211,"stars":212,"difficulty_score":51,"last_commit_at":213,"category_tags":214,"status":52},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",[40,196,197],{"id":216,"name":217,"github_repo":218,"description_zh":219,"stars":220,"difficulty_score":51,"last_commit_at":221,"category_tags":222,"status":52},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",[40,206],{"id":224,"name":225,"github_repo":226,"description_zh":227,"stars":228,"difficulty_score":51,"last_commit_at":229,"category_tags":230,"status":52},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",[196,231,232,41,197,233,206,40,234],"数据工具","视频","其他","音频",{"id":236,"name":237,"github_repo":238,"description_zh":239,"stars":240,"difficulty_score":193,"last_commit_at":241,"category_tags":242,"status":52},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",[197,196,40,206,233]]