[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-apache--singa":3,"tool-apache--singa":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",140436,2,"2026-04-05T23:32:43",[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":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":79,"owner_email":79,"owner_twitter":79,"owner_website":80,"owner_url":81,"languages":82,"stars":117,"forks":118,"last_commit_at":119,"license":120,"difficulty_score":10,"env_os":121,"env_gpu":122,"env_ram":122,"env_deps":123,"category_tags":126,"github_topics":127,"view_count":10,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":129,"updated_at":130,"faqs":131,"releases":167},979,"apache\u002Fsinga","singa","a distributed deep learning platform","SINGA 是一个开源的分布式深度学习平台，由 Apache 软件基金会支持。它专为高效训练和部署大规模深度学习模型而设计，通过将计算任务分散到多台机器上并行处理，解决了单机资源有限导致的训练速度慢、无法处理海量数据的问题。尤其适合开发者和研究人员使用——如果你需要构建复杂的神经网络（如图像识别或自然语言处理模型），或者面对超大规模数据集，SINGA 能显著提升训练效率，避免硬件瓶颈。它的技术亮点在于灵活的架构设计：同时支持 C++ 和 Python 接口，便于快速开发；内置自动并行化机制，简化分布式配置；还提供 Docker 容器化部署和详尽的测试覆盖，确保稳定性和易用性。作为社区驱动的项目，SINGA 拥有活跃的开发者生态和免费文档，让技术探索更轻松。无论你是算法工程师还是学术研究者，都能借助它更专注于模型创新，而非底层基础设施。","\u003C!--\n    Licensed to the Apache Software Foundation (ASF) under one\n    or more contributor license agreements.  See the NOTICE file\n    distributed with \u003C this work for additional information\n    regarding copyright ownership.  The ASF licenses this file\n    to you under the Apache License, Version 2.0 (the\n    \"License\"); you may not use this file except in compliance\n    with the License.  You may obtain a copy of the License at\n\n      http:\u002F\u002Fwww.apache.org\u002Flicenses\u002FLICENSE-2.0\n\n    Unless required by applicable law or agreed to in writing,\n    software distributed under the License is distributed on an\n    \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n    KIND, either express or implied.  See the License for the\n    specific language governing permissions and limitations\n    under the License.\n-->\n\n![Logo](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fapache_singa_readme_8b9bdd71141b.png)\n\n# Apache SINGA\n\n![Native Ubuntu build status](https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Fworkflows\u002FNative-Ubuntu\u002Fbadge.svg)\n![Native Mac build status](https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Fworkflows\u002FNative-MacOS\u002Fbadge.svg)\n![conda build status](https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Fworkflows\u002Fconda\u002Fbadge.svg)\n[![Documentation Status](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fapache_singa_readme_6bf48b3e9a6d.png)](https:\u002F\u002Fapache-singa.readthedocs.io\u002Fen\u002Flatest\u002F?badge=latest)\n![License](http:\u002F\u002Fimg.shields.io\u002F:license-Apache%202.0-blue.svg)\n[![Follow Apache SINGA on Twitter](https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Ffollow\u002Fapachesinga.svg?style=social&label=Follow)](https:\u002F\u002Ftwitter.com\u002FApacheSinga)\n[![Docker pulls](https:\u002F\u002Fimg.shields.io\u002Fdocker\u002Fpulls\u002Fapache\u002Fsinga.svg)](https:\u002F\u002Fhub.docker.com\u002Fr\u002Fapache\u002Fsinga\u002F)\n\nDistributed deep learning system\n\n[http:\u002F\u002Fsinga.apache.org](http:\u002F\u002Fsinga.apache.org)\n\n## Quick Start\n\n* [Installation](http:\u002F\u002Fsinga.apache.org\u002Fdocs\u002Finstallation\u002F)\n* [Examples](examples)\n\n## Issues\n\n* [JIRA tickets](https:\u002F\u002Fissues.apache.org\u002Fjira\u002Fbrowse\u002FSINGA)\n\n## Code Analysis:\n\n![LGTM C++ Grade](https:\u002F\u002Fimg.shields.io\u002Flgtm\u002Fgrade\u002Fcpp\u002Fgithub\u002Fapache\u002Fsinga)\n![LGTM Python Grade](https:\u002F\u002Fimg.shields.io\u002Flgtm\u002Fgrade\u002Fpython\u002Fgithub\u002Fapache\u002Fsinga)\n[![codecov](https:\u002F\u002Fcodecov.io\u002Fgh\u002Fapache\u002Fsinga\u002Fbranch\u002Fmaster\u002Fgraph\u002Fbadge.svg)](https:\u002F\u002Fcodecov.io\u002Fgh\u002Fapache\u002Fsinga)\n\n[![Stargazers over time](https:\u002F\u002Fstarchart.cc\u002Fapache\u002Fsinga.svg)](https:\u002F\u002Fstarchart.cc\u002Fapache\u002Fsinga)\n\n## Mailing Lists\n\n* [Development Mailing List](mailto:dev-subscribe@singa.apache.org) ([Archive](http:\u002F\u002Fmail-archives.apache.org\u002Fmod_mbox\u002Fsinga-dev\u002F))\n* [Commits Mailing List](mailto:commits-subscribe@singa.apache.org) ([Archive](http:\u002F\u002Fmail-archives.apache.org\u002Fmod_mbox\u002Fsinga-commits\u002F))\n","\u003C!--\n    Licensed to the Apache Software Foundation (ASF) under one\n    or more contributor license agreements.  See the NOTICE file\n    distributed with \u003C this work for additional information\n    regarding copyright ownership.  The ASF licenses this file\n    to you under the Apache License, Version 2.0 (the\n    \"License\"); you may not use this file except in compliance\n    with the License.  You may obtain a copy of the License at\n\n      http:\u002F\u002Fwww.apache.org\u002Flicenses\u002FLICENSE-2.0\n\n    Unless required by applicable law or agreed to in writing,\n    software distributed under the License is distributed on an\n    \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n    KIND, either express or implied.  See the License for the\n    specific language governing permissions and limitations\n    under the License.\n-->\n\n![Logo](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fapache_singa_readme_8b9bdd71141b.png)\n\n# Apache SINGA\n\n![Ubuntu原生构建状态](https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Fworkflows\u002FNative-Ubuntu\u002Fbadge.svg)\n![Mac原生构建状态](https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Fworkflows\u002FNative-MacOS\u002Fbadge.svg)\n![conda构建状态](https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Fworkflows\u002Fconda\u002Fbadge.svg)\n[![文档状态](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fapache_singa_readme_6bf48b3e9a6d.png)](https:\u002F\u002Fapache-singa.readthedocs.io\u002Fen\u002Flatest\u002F?badge=latest)\n![许可证](http:\u002F\u002Fimg.shields.io\u002F:license-Apache%202.0-blue.svg)\n[![在Twitter上关注Apache SINGA](https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Ffollow\u002Fapachesinga.svg?style=social&label=Follow)](https:\u002F\u002Ftwitter.com\u002FApacheSinga)\n[![Docker拉取次数](https:\u002F\u002Fimg.shields.io\u002Fdocker\u002Fpulls\u002Fapache\u002Fsinga.svg)](https:\u002F\u002Fhub.docker.com\u002Fr\u002Fapache\u002Fsinga\u002F)\n\n分布式深度学习 (deep learning) 系统\n\n[http:\u002F\u002Fsinga.apache.org](http:\u002F\u002Fsinga.apache.org)\n\n## 快速开始\n\n* [安装](http:\u002F\u002Fsinga.apache.org\u002Fdocs\u002Finstallation\u002F)\n* [示例](examples)\n\n## 问题\n\n* [JIRA问题](https:\u002F\u002Fissues.apache.org\u002Fjira\u002Fbrowse\u002FSINGA)\n\n## 代码分析：\n\n![LGTM C++评级](https:\u002F\u002Fimg.shields.io\u002Flgtm\u002Fgrade\u002Fcpp\u002Fgithub\u002Fapache\u002Fsinga)\n![LGTM Python评级](https:\u002F\u002Fimg.shields.io\u002Flgtm\u002Fgrade\u002Fpython\u002Fgithub\u002Fapache\u002Fsinga)\n[![代码覆盖率](https:\u002F\u002Fcodecov.io\u002Fgh\u002Fapache\u002Fsinga\u002Fbranch\u002Fmaster\u002Fgraph\u002Fbadge.svg)](https:\u002F\u002Fcodecov.io\u002Fgh\u002Fapache\u002Fsinga)\n\n[![星标随时间变化](https:\u002F\u002Fstarchart.cc\u002Fapache\u002Fsinga.svg)](https:\u002F\u002Fstarchart.cc\u002Fapache\u002Fsinga)\n\n## 邮件列表\n\n* [开发邮件列表](mailto:dev-subscribe@singa.apache.org) ([存档](http:\u002F\u002Fmail-archives.apache.org\u002Fmod_mbox\u002Fsinga-dev\u002F))\n* [提交邮件列表](mailto:commits-subscribe@singa.apache.org) ([存档](http:\u002F\u002Fmail-archives.apache.org\u002Fmod_mbox\u002Fsinga-commits\u002F))","# Apache SINGA 快速上手指南\n\n## 环境准备\n- **系统要求**：Ubuntu 18.04\u002F20.04 或 macOS 10.15+（64位系统）\n- **前置依赖**：\n  - Python 3.6+\n  - 基础构建工具：CMake、GCC、Make\n  - GPU支持需额外安装：NVIDIA驱动（450+）、CUDA 10.2+、cuDNN 7.6+\n  - 内存建议：≥8GB（编译时需充足内存）\n\n## 安装步骤\n推荐使用国内镜像加速安装（优先选择以下方案）：\n\n### 方案1：Docker安装（最快捷）\n```bash\n# 使用阿里云镜像加速（国内推荐）\ndocker pull registry.cn-hangzhou.aliyuncs.com\u002Fapache_singa\u002Fsinga:latest\n# 启动容器（含GPU支持）\nnvidia-docker run -it registry.cn-hangzhou.aliyuncs.com\u002Fapache_singa\u002Fsinga:latest\n```\n\n### 方案2：Conda安装（适合Python环境）\n```bash\n# 配置清华conda镜像（国内加速）\nconda config --add channels https:\u002F\u002Fmirrors.tuna.tsinghua.edu.cn\u002Fanaconda\u002Fcloud\u002Fconda-forge\u002F\nconda config --set show_channel_urls yes\n# 安装SINGA\nconda install -c conda-forge singa\n```\n\n### 方案3：源码编译（高级用户）\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga.git\ncd singa\nmkdir build && cd build\n# 使用清华源加速依赖下载（可选）\nexport PIP_INDEX_URL=https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\ncmake .. -G \"Unix Makefiles\"\nmake -j$(nproc)\n```\n\n## 基本使用\n运行最简张量操作示例（验证安装成功）：\n\n```python\nfrom singa import tensor\nimport numpy as np\n\n# 创建2x2全1张量\nx = tensor.Tensor((2, 2))\nx.set_value(1.0)\n\n# 执行加法运算\ny = x + x\n\n# 转换为NumPy数组并打印\nprint(y.to_numpy())\n# 预期输出：[[2. 2.] [2. 2.]]\n```\n\n> **提示**：完整训练示例请参考[官方examples目录](https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Ftree\u002Fmaster\u002Fexamples)，包含MNIST分类等基础任务。首次运行建议从`examples\u002Fvanilla_mnist`开始。","某知名电商平台的机器学习团队正开发新一代用户行为预测模型，需训练包含10亿参数的深度神经网络，处理每日10亿条用户点击流数据以优化推荐系统。\n\n### 没有 singa 时\n- 训练耗时过长：单机训练需72小时以上，模型迭代周期拖累产品上线节奏\n- 内存瓶颈突出：模型参数超出单GPU容量，频繁触发OOM错误，需手动拆分数据\n- 扩展配置复杂：添加新服务器需重写通信代码，多机同步易出错且耗时数天\n- 容错能力缺失：任意节点宕机即中断训练，平均每周损失10小时计算资源\n- 开发效率低下：工程师需编写大量底层分布式逻辑，50%精力用于调试而非模型优化\n\n### 使用 singa 后\n- 训练效率跃升：分布式框架自动分发任务，训练时间压缩至8小时内，支持每日模型迭代\n- 内存管理智能：自动模型分片技术充分利用集群资源，彻底规避OOM问题\n- 扩展操作简化：通过配置文件动态增减计算节点，5分钟内完成集群扩容\n- 容错机制可靠：自动保存检查点，节点故障后10分钟内恢复训练，资源损失趋近于零\n- 开发体验优化：高层API封装分布式细节，工程师专注模型设计，代码量减少40%\n\nsinga让大规模深度学习训练从繁琐运维转变为高效可靠的工程实践，显著加速AI产品落地。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fapache_singa_8b9bdd71.png","apache","The Apache Software Foundation","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fapache_c52803a1.png","",null,"https:\u002F\u002Fwww.apache.org\u002F","https:\u002F\u002Fgithub.com\u002Fapache",[83,87,91,95,99,102,105,109,113],{"name":84,"color":85,"percentage":86},"C++","#f34b7d",70.2,{"name":88,"color":89,"percentage":90},"Python","#3572A5",21,{"name":92,"color":93,"percentage":94},"C","#555555",4.8,{"name":96,"color":97,"percentage":98},"Cuda","#3A4E3A",1.1,{"name":100,"color":79,"percentage":101},"SWIG",1,{"name":103,"color":104,"percentage":101},"CMake","#DA3434",{"name":106,"color":107,"percentage":108},"Dockerfile","#384d54",0.6,{"name":110,"color":111,"percentage":112},"Shell","#89e051",0.4,{"name":114,"color":115,"percentage":116},"Java","#b07219",0.1,3591,1263,"2026-04-04T13:48:51","Apache-2.0","Linux, macOS","未说明",{"notes":124,"python":122,"dependencies":125},"支持通过 conda 和 Docker 安装部署，建议参考官方安装文档获取详细环境配置；README 中未明确指定具体依赖库和硬件要求。",[],[13],[128],"deep-learning","2026-03-27T02:49:30.150509","2026-04-06T08:25:32.722709",[132,137,141,145,149,154,159,163],{"id":133,"question_zh":134,"answer_zh":135,"source_url":136},4350,"如何保存模型的状态（如权重和运行时统计量）？","使用 Module 类的 save_states 方法。例如：`m1.save_states('.\u002Fsaved_models\u002Fmy_checkpoint_1')`。这会保存模型的状态（包括权重、BatchNorm 的运行均值和方差等），但不保存模型结构。状态文件可用于后续恢复训练或推理。","https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Fissues\u002F691",{"id":138,"question_zh":139,"answer_zh":140,"source_url":136},4351,"如何加载模型的状态以恢复训练或推理？","使用 Module 类的 load_states 方法。例如：`m1.load_states('.\u002Fsaved_models\u002Fmy_checkpoint_1', dev)`，其中 dev 是目标设备。这会将保存的状态（如权重）加载到模型中，并自动处理设备迁移。加载后需调用 compile 方法配置训练环境。",{"id":142,"question_zh":143,"answer_zh":144,"source_url":136},4352,"如何保存整个模型（包括结构和状态）为 ONNX 格式？","使用 singa.save 函数。例如：`singa.save('.\u002Fsaved_models\u002Fmy_model_1', m1)`。这会将模型结构和状态保存为 ONNX 文件，便于跨平台部署。加载时需配合 singa.load 和自定义模型类使用。",{"id":146,"question_zh":147,"answer_zh":148,"source_url":136},4353,"如何从 ONNX 文件加载模型并进行训练？","首先定义自定义模型类继承 SONNXModel，然后加载 ONNX 文件。例如：\n```python\nclass MySONNXModel(SONNXModel):\n    pass\nonnx_model = onnx.load('.\u002Fsaved_models\u002Fonnx_model_downloaded')\nm1 = MySONNXModel(onnx_model)\nm1.compile([placeholder_x], is_train=True, use_graph=True, graph_alg='default')\nfor _ in data:\n    m1.train_one_batch(_)\n```",{"id":150,"question_zh":151,"answer_zh":152,"source_url":153},4354,"Module.compile 方法的作用和正确用法是什么？","compile 方法用于设置训练配置、构建计算图并初始化参数。用法：`model.compile([tx], is_train=True, use_graph=graph, sequential=sequential)`。它会自动调用 forward 方法开启图模式，处理输入张量并配置训练环境。必须在训练前调用，否则无法正确构建计算图。","https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Fissues\u002F696",{"id":155,"question_zh":156,"answer_zh":157,"source_url":158},4355,"Layer 构造函数中参数初始化的最佳实践是什么？","参数初始化应在 __init__ 方法中完成，通过指定输入大小和初始化器。例如 Linear 层需用户提供输入尺寸以创建权重张量。避免在 call 函数中初始化，否则可能导致参数未正确创建或图构建失败。长期方案（V4.0）将统一移至 Layer 的 init 函数处理。","https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Fissues\u002F674",{"id":160,"question_zh":161,"answer_zh":162,"source_url":153},4356,"Autograd 模块中 Operator 和 Layer 的关系如何设计？","Operator 实现无状态的前向\u002F后向计算，Layer 存储状态（如参数和句柄）并调用 Operator。重构后建议统一使用 Layer 实例处理所有操作（包括 Flatten 等无状态操作），避免混合使用 Operator 函数。例如，每个 Operator 应有对应的 Layer 类，确保 API 一致性。",{"id":164,"question_zh":165,"answer_zh":166,"source_url":136},4357,"如何同时保存模型状态和自定义检查点数据（如 epoch ID）？","在 save 方法中通过 ckp_states 参数传递。例如：`m1.save('.\u002Fsaved_models\u002Fmy_model', ckp_states={'epoch': 10})`。这会将模型状态与自定义数据（如训练轮次）一并保存。加载时使用 load 方法返回的字典恢复这些状态。",[168,173,178],{"id":169,"version":170,"summary_zh":171,"released_at":172},113465,"3.0.0","Release note is [here](https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga\u002Fblob\u002Fmaster\u002FRELEASE_NOTES)","2020-04-21T08:01:08",{"id":174,"version":175,"summary_zh":176,"released_at":177},113466,"3.0.0.rc1","This release includes following changes:\r\n\r\n  * Code quality has been promoted by introducing linting check in CI and auto code formatter. \r\n    For linting, the tools, `cpplint` and `pylint`, are used and configured to comply \r\n    [google coding styles](http:\u002F\u002Fgoogle.github.io\u002Fstyleguide\u002F)  details in `tool\u002Flinting\u002F`. \r\n    Similarly, formatting tools, `clang-format` and `yapf` configured with google coding styles, \r\n    are the recommended one for developers to clean code before submitting changes, \r\n    details in `tool\u002Fcode-format\u002F`. [LGTM](https:\u002F\u002Flgtm.com) is enabled on Github for \r\n    code quality check; License check is also enabled.\r\n\r\n * New Tensor APIs are added for naming consistency, and feature enhancement: \r\n   - size(), mem_size(), get_value(), to_proto(), l1(), l2(): added for the sake of naming consistency\r\n   - AsType(): convert data type between `float` and `int`\r\n   - ceil(): perform element-wise ceiling of the input\r\n   - concat(): concatenate two tensor\r\n   - index selector: e.g. tensor1[:,:,1:,1:]\r\n   - softmax(in, axis): allow to perform softmax on a axis on a multi-dimensional tensor\r\n\r\n  * 14 new operators are added into the autograd module: Gemm, GlobalAveragePool, ConstantOfShape, \r\n    Dropout, ReduceSum, ReduceMean, Slice, Ceil, Split, Gather, Tile, NonZero, Cast, OneHot. \r\n    Their unit tests are added as well.\r\n\r\n  * 14 new operators are added to sonnx module for both backend and frontend: \r\n    [Gemm](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#Gemm), \r\n    [GlobalAveragePool](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#GlobalAveragePool), \r\n    [ConstantOfShape](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#ConstantOfShape), \r\n    [Dropout](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#Dropout), \r\n    [ReduceSum](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#ReduceSum), \r\n    [ReduceMean](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#ReduceMean), \r\n    [Slice](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#Slice), \r\n    [Ceil](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#Ceil), \r\n    [Split](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#Split), \r\n    [Gather](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#Gather), \r\n    [Tile](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#Tile), \r\n    [NonZero](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#NonZero), \r\n    [Cast](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#Cast), \r\n    [OneHot](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#OneHot). \r\n    Their tests are added as well.\r\n\r\n  * Some ONNX models are imported into SINGA, including \r\n    [Bert-squad](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fmodels\u002Ftree\u002Fmaster\u002Ftext\u002Fmachine_comprehension\u002Fbert-squad), \r\n    [Arcface](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fmodels\u002Ftree\u002Fmaster\u002Fvision\u002Fbody_analysis\u002Farcface), \r\n    [FER+ Emotion](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fmodels\u002Ftree\u002Fmaster\u002Fvision\u002Fbody_analysis\u002Femotion_ferplus), \r\n    [MobileNet](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fmodels\u002Ftree\u002Fmaster\u002Fvision\u002Fclassification\u002Fmobilenet), \r\n    [ResNet18](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fmodels\u002Ftree\u002Fmaster\u002Fvision\u002Fclassification\u002Fresnet), \r\n    [Tiny Yolov2](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fmodels\u002Ftree\u002Fmaster\u002Fvision\u002Fobject_detection_segmentation\u002Ftiny_yolov2), \r\n    [Vgg16](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fmodels\u002Ftree\u002Fmaster\u002Fvision\u002Fclassification\u002Fvgg), and Mnist.\r\n\r\n  * Some operators now support [multidirectional broadcasting](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FBroadcasting.md#multidirectional-broadcasting), \r\n    including Add, Sub, Mul, Div, Pow, PRelu, Gemm \r\n\r\n  * [Distributed training with communication optimization]. [DistOpt](.\u002Fpython\u002Fsinga\u002Fopt.py) \r\n    has implemented multiple optimization techniques, including gradient sparsification, \r\n    chunk transmission, and gradient compression.\r\n\r\n  * Computational graph construction at the CPP level. The operations submitted to the Device are buffered.\r\n    After analyzing the dependency, the computational graph is created, which is further analyzed for\r\n    speed and memory optimization. To enable this feature, use the [Module API](.\u002Fpython\u002Fsinga\u002Fmodule.py).\r\n\r\n  * New website based on Docusaurus. The documentation files are moved to a separate repo [singa-doc]](https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga-doc).\r\n    The static website files are stored at [singa-site](https:\u002F\u002Fgithub.com\u002Fapache\u002Fsinga-site).\r\n\r\n  * DNNL([Deep Neural Network Library](https:\u002F\u002Fgithub.com\u002Fintel\u002Fmkl-dnn)), powered by Intel, \r\n    is integrated into `model\u002Foperations\u002F[batchnorm|pooling|convolution]`, \r\n    the changes is opaque to the end users. The current version is dnnl v1.1 \r\n    which replaced previous integration of mkl-dnn v0.18. The framework could \r\n    boost the performance of dl operations when executing on CPU. The dnnl dependency \r\n    is installed through conda.\r\n\r\n  * Some Tensor APIs are marked as deprecated which could be replaced by broadcast, \r\n    an","2020-04-08T17:24:47",{"id":179,"version":180,"summary_zh":181,"released_at":182},113467,"2.0.0","In this release, we have added the support of ONNX, implemented the CPU operations using MKLDNN, added operations for autograd, and updated the dependent libraries and CI tools.\r\n\r\n* Core components\r\n    * [SINGA-434] Support tensor broadcasting\r\n    * [SINGA-370] Improvement to tensor reshape and various misc. changes related to SINGA-341 and 351\r\n\r\n* Model components\r\n    * [SINGA-333] Add support for Open Neural Network Exchange (ONNX) format\r\n    * [SINGA-385] Add new python module for optimizers\r\n    * [SINGA-394] Improve the CPP operations via Intel MKL DNN lib\r\n    * [SINGA-425] Add 3 operators , Abs(), Exp() and leakyrelu(), for Autograd \r\n    * [SINGA-410] Add two function, set_params() and get_params(), for Autograd Layer class\r\n    * [SINGA-383] Add Separable Convolution for autograd\r\n    * [SINGA-388] Develop some RNN layers by calling tiny operations like matmul, addbias.\r\n    * [SINGA-382] Implement concat operation for autograd    \r\n    * [SINGA-378] Implement maxpooling operation and its related functions for autograd\r\n    * [SINGA-379] Implement batchnorm operation and its related functions for autograd\r\n\r\n* Utility functions and CI\r\n    * [SINGA-432] Update depdent lib versions in conda-build config\r\n    * [SINGA-429] Update docker images for latest cuda and cudnn\r\n    * [SINGA-428] Move Docker images under Apache user name\r\n\r\n* Documentation and usability\r\n    * [SINGA-395] Add documentation for autograd APIs\r\n    * [SINGA-344] Add a GAN example\r\n    * [SINGA-390] Update installation.md\r\n    * [SINGA-384] Implement ResNet using autograd API\r\n    * [SINGA-352] Complete SINGA documentation in Chinese version\r\n      \r\n* Bugs fixed\r\n    * [SINGA-431] Unit Test failed - Tensor Transpose\r\n    * [SINGA-422] ModuleNotFoundError: No module named \"_singa_wrap\"\r\n    * [SINGA-418] Unsupportive type 'long' in python3.  \r\n    * [SINGA-409] Basic `singa-cpu` import throws error\r\n    * [SINGA-408] Unsupportive function definition in python3\r\n    * [SINGA-380] Fix bugs from Reshape ","2019-04-07T06:32:58"]