[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-onnx--onnx":3,"tool-onnx--onnx":62},[4,18,26,36,46,54],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":10,"last_commit_at":24,"category_tags":25,"status":17},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,"2026-04-05T11:01:52",[14,15,13],{"id":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":32,"last_commit_at":33,"category_tags":34,"status":17},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",160784,2,"2026-04-19T11:32:54",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":42,"last_commit_at":43,"category_tags":44,"status":17},8272,"opencode","anomalyco\u002Fopencode","OpenCode 是一款开源的 AI 编程助手（Coding Agent），旨在像一位智能搭档一样融入您的开发流程。它不仅仅是一个代码补全插件，而是一个能够理解项目上下文、自主规划任务并执行复杂编码操作的智能体。无论是生成全新功能、重构现有代码，还是排查难以定位的 Bug，OpenCode 都能通过自然语言交互高效完成，显著减少开发者在重复性劳动和上下文切换上的时间消耗。\n\n这款工具专为软件开发者、工程师及技术研究人员设计，特别适合希望利用大模型能力来提升编码效率、加速原型开发或处理遗留代码维护的专业人群。其核心亮点在于完全开源的架构，这意味着用户可以审查代码逻辑、自定义行为策略，甚至私有化部署以保障数据安全，彻底打破了传统闭源 AI 助手的“黑盒”限制。\n\n在技术体验上，OpenCode 提供了灵活的终端界面（Terminal UI）和正在测试中的桌面应用程序，支持 macOS、Windows 及 Linux 全平台。它兼容多种包管理工具，安装便捷，并能无缝集成到现有的开发环境中。无论您是追求极致控制权的资深极客，还是渴望提升产出的独立开发者，OpenCode 都提供了一个透明、可信",144296,1,"2026-04-16T14:50:03",[13,45],"插件",{"id":47,"name":48,"github_repo":49,"description_zh":50,"stars":51,"difficulty_score":32,"last_commit_at":52,"category_tags":53,"status":17},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。",109154,"2026-04-18T11:18:24",[14,15,13],{"id":55,"name":56,"github_repo":57,"description_zh":58,"stars":59,"difficulty_score":32,"last_commit_at":60,"category_tags":61,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[45,13,15,14],{"id":63,"github_repo":64,"name":65,"description_en":66,"description_zh":67,"ai_summary_zh":67,"readme_en":68,"readme_zh":69,"quickstart_zh":70,"use_case_zh":71,"hero_image_url":72,"owner_login":65,"owner_name":73,"owner_avatar_url":74,"owner_bio":75,"owner_company":76,"owner_location":76,"owner_email":76,"owner_twitter":76,"owner_website":77,"owner_url":78,"languages":79,"stars":105,"forks":106,"last_commit_at":107,"license":108,"difficulty_score":42,"env_os":109,"env_gpu":110,"env_ram":110,"env_deps":111,"category_tags":117,"github_topics":118,"view_count":32,"oss_zip_url":76,"oss_zip_packed_at":76,"status":17,"created_at":131,"updated_at":132,"faqs":133,"releases":163},9772,"onnx\u002Fonnx","onnx","Open standard for machine learning interoperability","ONNX（Open Neural Network Exchange）是一个开放的机器学习互操作标准，旨在为人工智能开发者提供一个统一的模型格式。它就像 AI 领域的“通用翻译器”，允许用户将深度学习或传统机器学习模型在不同框架（如 PyTorch、TensorFlow）和硬件平台之间自由转换与部署，从而打破技术生态间的壁垒。\n\n在 AI 开发中，研究人员常面临训练框架与推理环境不兼容的难题，导致模型从实验到落地的过程繁琐且低效。ONNX 通过定义可扩展的计算图模型、内置算子及标准数据类型，完美解决了这一痛点，让模型能够“一次导出，到处运行”，显著加速了从科研探索到生产应用的创新周期。\n\n这款工具非常适合 AI 开发者、算法研究人员以及需要跨平台部署模型的工程师使用。无论是希望灵活切换开发工具的研究者，还是致力于优化推理性能的系统架构师，都能从中受益。\n\nONNX 的核心亮点在于其强大的生态系统兼容性，目前已被众多主流框架和硬件厂商广泛支持。它不仅专注于高效的推理能力，还提供了完善的图形优化、版本转换及形状推断等实用工具，帮助社区共同推动人工智能技术的开放与演进。","\u003C!--\nCopyright (c) ONNX Project Contributors\n\nSPDX-License-Identifier: Apache-2.0\n-->\n\n\u003Cp align=\"center\">\u003Cimg width=\"40%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fonnx_onnx_readme_a5b8734a828e.png\" \u002F>\u003C\u002Fp>\n\n[![PyPI - Version](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fonnx.svg)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fonnx)\n[![CI](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Factions\u002Fworkflows\u002Fmain.yml\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Factions\u002Fworkflows\u002Fmain.yml)\n[![CII Best Practices](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fonnx_onnx_readme_eda489f85045.png)](https:\u002F\u002Fbestpractices.coreinfrastructure.org\u002Fprojects\u002F3313)\n[![OpenSSF Scorecard](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fonnx_onnx_readme_66c6bba21077.png)](https:\u002F\u002Fapi.securityscorecards.dev\u002Fprojects\u002Fgithub.com\u002Fonnx\u002Fonnx)\n[![REUSE compliant](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fonnx_onnx_readme_42921db67a80.png)](https:\u002F\u002Fapi.reuse.software\u002Finfo\u002Fgithub.com\u002Fonnx\u002Fonnx)\n[![Ruff](https:\u002F\u002Fimg.shields.io\u002Fendpoint?url=https:\u002F\u002Fraw.githubusercontent.com\u002Fastral-sh\u002Fruff\u002Fmain\u002Fassets\u002Fbadge\u002Fv2.json)](https:\u002F\u002Fgithub.com\u002Fastral-sh\u002Fruff)\n[![abi3 compatible](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fabi3-compatible-brightgreen)](https:\u002F\u002Fdocs.python.org\u002F3\u002Fc-api\u002Fstable.html)\n\n[Open Neural Network Exchange (ONNX)](https:\u002F\u002Fonnx.ai) is an open ecosystem that empowers AI developers\nto choose the right tools as their project evolves. ONNX provides an open source format for AI models, both deep learning and traditional ML. It defines an extensible computation graph model, as well as definitions of built-in operators and standard\ndata types. Currently we focus on the capabilities needed for inferencing (scoring).\n\nONNX is [widely supported](http:\u002F\u002Fonnx.ai\u002Fsupported-tools) and can be found in many frameworks, tools, and hardware. Enabling interoperability between different frameworks and streamlining the path from research to production helps increase the speed of innovation in the AI community. We invite the community to join us and further evolve ONNX.\n\n\n# Use ONNX\n\n* [Documentation of ONNX Python Package](https:\u002F\u002Fonnx.ai\u002Fonnx\u002F)\n* [Tutorials for creating ONNX models](https:\u002F\u002Fgithub.com\u002Fonnx\u002Ftutorials)\n* [Pre-trained ONNX models](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fmodels)\n\n# Learn about the ONNX spec\n\n* [Overview](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOverview.md)\n* [ONNX intermediate representation spec](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FIR.md)\n* [Versioning principles of the spec](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FVersioning.md)\n* [Operators documentation](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md)\n* [Operators documentation](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Findex.html) (latest release)\n* [Python API Overview](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FPythonAPIOverview.md)\n\n# Programming utilities for working with ONNX Graphs\n\n* [Shape and Type Inference](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FShapeInference.md)\n* [Graph Optimization](https:\u002F\u002Fgithub.com\u002Fonnx\u002Foptimizer)\n* [Opset Version Conversion](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002Fdocsgen\u002Fsource\u002Fapi\u002Fversion_converter.md)\n\n# Contribute\n\nONNX is a community project and the open governance model is described [here](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fcommunity\u002Freadme.md). We encourage you to join the effort and contribute feedback, ideas, and code. You can participate in the [Special Interest Groups](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fcommunity\u002Fsigs.md) and [Working Groups](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fcommunity\u002Fworking-groups.md) to shape the future of ONNX.\n\nCheck out our [contribution guide](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002FCONTRIBUTING.md) to get started.\n\nIf you think some operator should be added to ONNX specification, please read\n[this document](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FAddNewOp.md).\n\n# Community meetings\n\nThe schedules of the regular meetings of the Steering Committee, the working groups and the SIGs can be found [here](https:\u002F\u002Fonnx.ai\u002Fcalendar)\n\nCommunity Meetups are held at least once a year. Content from previous community meetups are at:\n\n* 2020.04.09 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14091402\u002FLF+AI+Day+-ONNX+Community+Virtual+Meetup+-+Silicon+Valley+-+2020+April+9>\n* 2020.10.14 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14092138\u002FLF+AI+Day+-+ONNX+Community+Workshop+-+2020+October+14>\n* 2021.03.24 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14092424\u002FInstructions+for+Event+Hosts+-+LF+AI+Data+Day+-+ONNX+Virtual+Community+Meetup+-+March+2021>\n* 2021.10.21 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14093194\u002FLF+AI+Data+Day+ONNX+Community+Virtual+Meetup+-+October+2021>\n* 2022.06.24 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14093969\u002FONNX+Community+Day+-+2022+June+24>\n* 2023.06.28 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14094507\u002FONNX+Community+Day+2023+-+June+28>\n\n# Discuss\n\nWe encourage you to open [Issues](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues), or use [Slack](https:\u002F\u002Flfaifoundation.slack.com\u002F) (If you have not joined yet, please use this [link](https:\u002F\u002Fjoin.slack.com\u002Ft\u002Flfaifoundation\u002Fshared_invite\u002Fzt-o65errpw-gMTbwNr7FnNbVXNVFkmyNA) to join the group) for more real-time discussion.\n\n# Follow Us\n\nStay up to date with the latest ONNX news. [[Facebook](https:\u002F\u002Fwww.facebook.com\u002Fonnxai\u002F)] [[Twitter\u002FX](https:\u002F\u002Ftwitter.com\u002Fonnxai)]\n\n# Roadmap\n\nA roadmap process takes place every year. More details can be found [here](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fsteering-committee\u002Ftree\u002Fmain\u002Froadmap)\n\n# Installation\n\nONNX released packages are published in PyPi.\n\n```sh\npip install onnx # or pip install onnx[reference] for optional reference implementation dependencies\n```\n\n[ONNX weekly packages](https:\u002F\u002Fpypi.org\u002Fproject\u002Fonnx-weekly\u002F) are published in PyPI to enable experimentation and early testing.\n\nDetailed install instructions, including Common Build Options and Common Errors can be found [here](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002FINSTALL.md)\n\n# Python ABI3 Compatibility\n\nThis package provides [abi3](https:\u002F\u002Fdocs.python.org\u002F3\u002Fc-api\u002Fstable.html)-compatible wheels, allowing a single binary wheel to work across multiple Python versions (from 3.12 onwards).\n\n\n# Testing\n\nONNX uses [pytest](https:\u002F\u002Fdocs.pytest.org) as test driver. In order to run tests, you will first need to install `pytest`:\n\n```sh\npip install pytest\n```\n\nAfter installing pytest, use the following command to run tests.\n\n```sh\npytest\n```\n\n# Development\n\nCheck out the [contributor guide](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002FCONTRIBUTING.md) for instructions.\n\n# Reproducible Builds (Linux)\n\nThis project provides reproducible builds for Linux.\n\nA *reproducible build* means that the same source code will always produce identical binary outputs, no matter who builds it or where it is built.\n\nTo achieve this, we use the [`SOURCE_DATE_EPOCH`](https:\u002F\u002Freproducible-builds.org\u002Fdocs\u002Fsource-date-epoch\u002F) standard. This ensures that build timestamps and other time-dependent information are fixed, making the output bit-for-bit identical across different environments.\n\n### Why this matters\n- **Transparency**: Anyone can verify that the distributed binaries were created from the published source code.\n- **Security**: Prevents tampering or hidden changes in the build process.\n- **Trust**: Users can be confident that the binaries they download are exactly what the maintainers intended.\n\nIf you prefer, you can use the prebuilt reproducible binaries instead of building from source yourself.\n\n# License\n\n[Apache License v2.0](LICENSE)\n\n# Trademark\nCheckout [https:\u002F\u002Ftrademarks.justia.com](https:\u002F\u002Ftrademarks.justia.com\u002F877\u002F25\u002Fonnx-87725026.html) for the trademark.\n\n[General rules of the Linux Foundation on Trademark usage](https:\u002F\u002Fwww.linuxfoundation.org\u002Flegal\u002Ftrademark-usage)\n\n# Code of Conduct\n\n[ONNX Open Source Code of Conduct](https:\u002F\u002Fonnx.ai\u002Fcodeofconduct.html)\n","\u003C!--\n版权所有 © ONNX 项目贡献者\n\nSPDX 许可证标识：Apache-2.0\n-->\n\n\u003Cp align=\"center\">\u003Cimg width=\"40%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fonnx_onnx_readme_a5b8734a828e.png\" \u002F>\u003C\u002Fp>\n\n[![PyPI - 版本](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fonnx.svg)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fonnx)\n[![CI](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Factions\u002Fworkflows\u002Fmain.yml\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Factions\u002Fworkflows\u002Fmain.yml)\n[![CII 最佳实践](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fonnx_onnx_readme_eda489f85045.png)](https:\u002F\u002Fbestpractices.coreinfrastructure.org\u002Fprojects\u002F3313)\n[![OpenSSF 评分卡](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fonnx_onnx_readme_66c6bba21077.png)](https:\u002F\u002Fapi.securityscorecards.dev\u002Fprojects\u002Fgithub.com\u002Fonnx\u002Fonnx)\n[![REUSE 合规](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fonnx_onnx_readme_42921db67a80.png)](https:\u002F\u002Fapi.reuse.software\u002Finfo\u002Fgithub.com\u002Fonnx\u002Fonnx)\n[![Ruff](https:\u002F\u002Fimg.shields.io\u002Fendpoint?url=https:\u002F\u002Fraw.githubusercontent.com\u002Fastral-sh\u002Fruff\u002Fmain\u002Fassets\u002Fbadge\u002Fv2.json)](https:\u002F\u002Fgithub.com\u002Fastral-sh\u002Fruff)\n[![abi3 兼容](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fabi3-compatible-brightgreen)](https:\u002F\u002Fdocs.python.org\u002F3\u002Fc-api\u002Fstable.html)\n\n[开放神经网络交换格式（ONNX）](https:\u002F\u002Fonnx.ai) 是一个开放的生态系统，使 AI 开发者能够在项目演进过程中选择最适合的工具。ONNX 提供了一种面向人工智能模型的开源格式，涵盖深度学习和传统机器学习领域。它定义了一个可扩展的计算图模型，以及内置算子和标准数据类型的规范。目前，我们的重点是推理（打分）所需的功能。\n\nONNX 受到 [广泛支持](http:\u002F\u002Fonnx.ai\u002Fsupported-tools)，并已集成到众多框架、工具和硬件中。通过实现不同框架之间的互操作性，并简化从研究到生产的流程，ONNX 有助于加速 AI 社区的创新步伐。我们诚挚邀请社区成员加入我们，共同推动 ONNX 的进一步发展。\n\n\n# 使用 ONNX\n\n* [ONNX Python 包文档](https:\u002F\u002Fonnx.ai\u002Fonnx\u002F)\n* [创建 ONNX 模型的教程](https:\u002F\u002Fgithub.com\u002Fonnx\u002Ftutorials)\n* [预训练的 ONNX 模型](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fmodels)\n\n# 了解 ONNX 规范\n\n* [概述](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOverview.md)\n* [ONNX 中间表示规范](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FIR.md)\n* [规范的版本控制原则](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FVersioning.md)\n* [算子文档](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md)\n* [算子文档](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Findex.html)（最新版本）\n* [Python API 概述](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FPythonAPIOverview.md)\n\n# 用于处理 ONNX 图的编程工具\n\n* [形状与类型推断](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FShapeInference.md)\n* [图优化](https:\u002F\u002Fgithub.com\u002Fonnx\u002Foptimizer)\n* [Opset 版本转换](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002Fdocsgen\u002Fsource\u002Fapi\u002Fversion_converter.md)\n\n# 贡献\n\nONNX 是一个社区驱动的项目，其开放治理模式在 [这里](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fcommunity\u002Freadme.md) 有所描述。我们鼓励您参与其中，提供反馈、想法和代码。您可以加入 [特别兴趣小组](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fcommunity\u002Fsigs.md) 和 [工作组](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fcommunity\u002Fworking-groups.md)，共同塑造 ONNX 的未来。\n\n请参阅我们的 [贡献指南](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002FCONTRIBUTING.md)，开始您的贡献之旅。\n\n如果您认为某些算子应被添加到 ONNX 规范中，请阅读 [本文档](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FAddNewOp.md)。\n\n# 社区会议\n\n指导委员会、工作组和 SIG 的定期会议日程可在 [这里](https:\u002F\u002Fonnx.ai\u002Fcalendar) 查看。\n\n社区聚会每年至少举行一次。以往社区聚会的相关内容如下：\n\n* 2020年4月9日 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14091402\u002FLF+AI+Day+-ONNX+Community+Virtual+Meetup+-+Silicon+Valley+-+2020+April+9>\n* 2020年10月14日 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14092138\u002FLF+AI+Day+-+ONNX+Community+Workshop+-+2020+October+14>\n* 2021年3月24日 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14092424\u002FInstructions+for+Event+Hosts+-+LF+AI+Data+Day+-+ONNX+Virtual+Community+Meetup+-+March+2021>\n* 2021年10月21日 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14093194\u002FLF+AI+Data+Day+ONNX+Community+Virtual+Meetup+-+October+2021>\n* 2022年6月24日 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14093969\u002FONNX+Community+Day+-+2022+June+24>\n* 2023年6月28日 \u003Chttps:\u002F\u002Flf-aidata.atlassian.net\u002Fwiki\u002Fspaces\u002FDL\u002Fpages\u002F14094507\u002FONNX+Community+Day+2023+-+June+28>\n\n# 讨论\n\n我们鼓励您打开 [问题](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues)，或使用 [Slack](https:\u002F\u002Flfaifoundation.slack.com\u002F)（如果您尚未加入，请使用此 [链接](https:\u002F\u002Fjoin.slack.com\u002Ft\u002Flfaifoundation\u002Fshared_invite\u002Fzt-o65errpw-gMTbwNr7FnNbVXNVFkmyNA) 加入群组）进行更实时的讨论。\n\n# 关注我们\n\n及时了解最新的 ONNX 动态。[[Facebook](https:\u002F\u002Fwww.facebook.com\u002Fonnxai\u002F)] [[Twitter\u002FX](https:\u002F\u002Ftwitter.com\u002Fonnxai)]\n\n# 路线图\n\n每年都会制定路线图。更多详情请参见 [这里](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fsteering-committee\u002Ftree\u002Fmain\u002Froadmap)。\n\n# 安装\n\nONNX 发布的软件包已在 PyPi 上架。\n\n```sh\npip install onnx # 或 pip install onnx[reference] 以获取可选的参考实现依赖项\n```\n\n[ONNX 每周版](https:\u002F\u002Fpypi.org\u002Fproject\u002Fonnx-weekly\u002F) 也在 PyPI 上发布，以便于实验和早期测试。\n\n详细的安装说明，包括常见构建选项和常见错误，可在 [这里](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002FINSTALL.md) 找到。\n\n# Python ABI3 兼容性\n\n本包提供了 [abi3](https:\u002F\u002Fdocs.python.org\u002F3\u002Fc-api\u002Fstable.html)-兼容的轮子，允许单个二进制轮子在多个 Python 版本上运行（从 3.12 开始）。\n\n\n# 测试\n\nONNX 使用 [pytest](https:\u002F\u002Fdocs.pytest.org) 作为测试驱动程序。要运行测试，您需要先安装 `pytest`：\n\n```sh\npip install pytest\n```\n\n安装完 pytest 后，使用以下命令运行测试：\n\n```sh\npytest\n```\n\n# 开发\n\n有关说明，请参阅 [贡献者指南](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002FCONTRIBUTING.md)。\n\n# 可重复构建（Linux）\n\n该项目为 Linux 提供可重复构建功能。\n\n所谓“可重复构建”，是指无论由谁、在何处构建，相同的源代码始终会产生完全相同的二进制输出。\n\n为此，我们采用 [`SOURCE_DATE_EPOCH`](https:\u002F\u002Freproducible-builds.org\u002Fdocs\u002Fsource-date-epoch\u002F) 标准。这确保了构建时间戳及其他依赖于时间的信息被固定，从而使输出在不同环境中逐位相同。\n\n### 为什么这很重要\n- **透明性**：任何人都可以验证分发的二进制文件确实是由已发布的源代码构建而成。\n- **安全性**：防止在构建过程中被篡改或引入隐藏的更改。\n- **信任**：用户可以确信他们下载的二进制文件正是维护者所期望的版本。\n\n如果您愿意，也可以直接使用预先构建好的可复现二进制文件，而无需自行从源代码构建。\n\n# 许可证\n\n[Apache许可证v2.0](LICENSE)\n\n# 商标\n有关商标的信息，请访问 [https:\u002F\u002Ftrademarks.justia.com](https:\u002F\u002Ftrademarks.justia.com\u002F877\u002F25\u002Fonnx-87725026.html)。\n\n[Linux基金会关于商标使用的通用规则](https:\u002F\u002Fwww.linuxfoundation.org\u002Flegal\u002Ftrademark-usage)\n\n# 行为准则\n\n[ONNX开源行为准则](https:\u002F\u002Fonnx.ai\u002Fcodeofconduct.html)","# ONNX 快速上手指南\n\nONNX (Open Neural Network Exchange) 是一个开放的生态系统，旨在为 AI 模型（包括深度学习和传统机器学习）提供通用的开源格式。它定义了可扩展的计算图模型、内置算子和标准数据类型，主要专注于推理（Inferencing）能力，帮助开发者在不同框架和硬件之间实现互操作性。\n\n## 环境准备\n\n*   **操作系统**：支持 Linux, macOS, Windows。\n*   **Python 版本**：推荐 Python 3.8 及以上版本（支持 abi3 兼容轮子，适用于 Python 3.12+）。\n*   **前置依赖**：\n    *   `pip` 包管理工具\n    *   `protobuf` (安装 onnx 时通常会自动处理，若需手动编译则需预先安装)\n\n## 安装步骤\n\n### 1. 基础安装\n通过 PyPI 安装最新稳定版：\n\n```bash\npip install onnx\n```\n\n> **国内加速建议**：如果您在中国大陆，建议使用清华或阿里镜像源以加快下载速度：\n> ```bash\n> pip install onnx -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n> ```\n\n### 2. 可选依赖安装\n如果您需要参考实现（reference implementation）的相关依赖：\n\n```bash\npip install \"onnx[reference]\"\n```\n\n### 3. 验证安装\n安装完成后，可通过以下命令检查版本：\n\n```bash\npython -c \"import onnx; print(onnx.__version__)\"\n```\n\n## 基本使用\n\n以下是使用 ONNX Python API 加载模型并进行简单形状推断的最简示例。\n\n### 示例：加载模型并执行形状推断\n\n假设您已经有一个名为 `model.onnx` 的模型文件：\n\n```python\nimport onnx\nfrom onnx import shape_inference\n\n# 1. 加载现有的 ONNX 模型\nmodel = onnx.load(\"model.onnx\")\n\n# 2. 执行形状推断 (Shape Inference)\n# 这会根据输入数据的形状推导出图中所有中间张量的形状\ninferred_model = shape_inference.infer_shapes(model)\n\n# 3. 保存推断后的模型\nonnx.save(inferred_model, \"model_inferred.onnx\")\n\n# 4. 检查模型有效性\ntry:\n    onnx.checker.check_model(inferred_model)\n    print(\"模型检查通过！\")\nexcept onnx.checker.ValidationError as e:\n    print(f\"模型无效：{e}\")\n```\n\n### 示例：从零创建一个简单的计算图\n\n```python\nimport onnx\nfrom onnx import helper, TensorProto\n\n# 定义输入\nX = helper.make_tensor_value_info('X', TensorProto.FLOAT, [1, 3])\n# 定义输出\nY = helper.make_tensor_value_info('Y', TensorProto.FLOAT, [1, 3])\n\n# 定义节点 (例如：Identity 操作)\nnode_def = helper.make_node(\n    'Identity',\n    inputs=['X'],\n    outputs=['Y'],\n)\n\n# 创建图形\ngraph_def = helper.make_graph(\n    [node_def],\n    'test-model',\n    [X],\n    [Y],\n)\n\n# 创建模型\nmodel_def = helper.make_model(graph_def, producer_name='onnx-example')\n\n# 保存模型\nonnx.save(model_def, 'simple_model.onnx')\nprint(\"简单模型已创建并保存为 simple_model.onnx\")\n```\n\n更多详细教程和预训练模型请访问 [ONNX Tutorials](https:\u002F\u002Fgithub.com\u002Fonnx\u002Ftutorials) 和 [ONNX Models](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fmodels)。","某自动驾驶团队需要将算法工程师在 PyTorch 中研发的最新障碍物检测模型，快速部署到由不同供应商提供的车载嵌入式推理设备上。\n\n### 没有 onnx 时\n- **框架绑定严重**：模型被锁定在 PyTorch 生态中，而目标硬件厂商仅支持 TensorRT 或自研推理引擎，导致无法直接运行。\n- **重复开发成本高**：为了适配不同硬件，团队必须手动用 C++ 重写网络结构或寻找不稳定的第三方转换脚本，极易引入误差。\n- **迭代周期漫长**：每次算法更新都需要经历漫长的“重新训练 - 手动移植 - 调试对齐”过程，严重拖慢从实验室到路测的进度。\n- **维护难度极大**：多套代码库并行维护，一旦底层算子变更，所有适配版本都需同步修改，协作效率低下。\n\n### 使用 onnx 后\n- **实现无缝互通**：利用 onnx 将 PyTorch 模型导出为标准的中间格式，该格式能被 TensorRT、OpenVINO 等主流推理引擎直接读取，打破框架壁垒。\n- **标准化部署流程**：团队只需维护一份 onnx 模型文件，即可通过各硬件厂商提供的标准适配器一键转换为特定格式，无需手动重写代码。\n- **加速研发闭环**：算法迭代后仅需重新导出 onnx 文件，部署端自动更新，将模型上线时间从数天缩短至数小时。\n- **统一验证标准**：onnx 提供了标准的算子定义和形状推断工具，确保模型在不同平台间的计算逻辑一致，大幅降低调试成本。\n\nonnx 通过建立统一的模型交换标准，彻底消除了深度学习框架与推理硬件之间的隔阂，让 AI 创新能以最快速度落地生产。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fonnx_onnx_a5b8734a.png","Open Neural Network Exchange","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fonnx_e2e75b88.png","ONNX is an open ecosystem for interoperable AI models. It's a community project: we welcome your contributions!",null,"https:\u002F\u002Fonnx.ai","https:\u002F\u002Fgithub.com\u002Fonnx",[80,84,88,92,96,99,102],{"name":81,"color":82,"percentage":83},"Python","#3572A5",52.7,{"name":85,"color":86,"percentage":87},"C++","#f34b7d",46.6,{"name":89,"color":90,"percentage":91},"CMake","#DA3434",0.6,{"name":93,"color":94,"percentage":95},"Shell","#89e051",0,{"name":97,"color":98,"percentage":95},"C","#555555",{"name":100,"color":101,"percentage":95},"PowerShell","#012456",{"name":103,"color":104,"percentage":95},"Batchfile","#C1F12E",20684,3921,"2026-04-19T12:10:08","Apache-2.0","Linux, macOS, Windows","未说明",{"notes":112,"python":113,"dependencies":114},"ONNX 本身是一个模型格式标准和转换工具，主要用于推理和模型互操作，不直接依赖特定 GPU。可通过 'pip install onnx' 安装，可选安装参考实现依赖 'pip install onnx[reference]'。详细编译选项和常见错误请参考 INSTALL.md。项目提供 Linux 下的可复现构建。","3.8+ (支持 abi3 兼容，二进制包覆盖 Python 3.12 及以上)",[115,116],"protobuf","numpy",[15,13,14],[119,120,121,65,122,123,124,125,126,127,128,129,130],"deep-learning","deep-neural-networks","neural-network","pytorch","tensorflow","keras","scikit-learn","ml","machine-learning","dnn","ai","artificial-intelligence","2026-03-27T02:49:30.150509","2026-04-20T04:06:09.549366",[134,139,144,149,154,159],{"id":135,"question_zh":136,"answer_zh":137,"source_url":138},43879,"在 Windows 上导入 ONNX 时出现 'DLL load failed' 错误（特别是升级到 onnx 1.16.2 后），如何解决？","该问题通常是由于 Microsoft Visual C++ Redistributable 版本过旧导致的。解决方案是升级 MSVC Redistributable 到 14.38 或更高版本。\n具体步骤：\n1. 访问微软官网或通过包管理器安装最新的 VC++ 运行库。\n2. 如果使用 Chocolatey，可运行命令：choco install vcredist140\n3. 确保安装的是 x64 版本（如果是 64 位系统）。\n注意：对于 ARM64 架构的 Windows 设备，可能需要手动查找并复制 msvcp140.dll 文件，因为某些安装包可能未将其部署到系统目录。","https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F6267",{"id":140,"question_zh":141,"answer_zh":142,"source_url":143},43880,"在 Linux (如 Arch Linux) 上从源码构建 ONNX _wheel_ 失败，报错与 protobuf 相关，如何解决？","这通常是因为系统中缺少 protobuf 开发库或版本不兼容。\n解决方案：\n1. 对于 Arch Linux 用户，运行命令：pacman -S protobuf 安装系统级的 protobuf 库。\n2. 确保安装的 protobuf 版本满足 ONNX 的要求（较新版本可能需要 protobuf > 5 或特定版本如 v25.1+）。\n3. 如果仍然失败，尝试清理缓存后重新构建：pip cache purge 然后重新运行 pip install。","https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F4376",{"id":145,"question_zh":146,"answer_zh":147,"source_url":148},43881,"ONNX 对 protobuf 版本的最低要求是什么？遇到版本冲突怎么办？","ONNX 的较新版本（如构建自源码或使用最新 release）通常要求 protobuf 版本高于 5.0（例如 v25.1+）。\n如果遇到类似 'Minimum protobuf version is upgraded' 的提示或运行时错误：\n1. 升级 protobuf：pip install --upgrade protobuf\n2. 检查当前版本：pip show protobuf\n3. 如果项目依赖固定了旧版本，需更新项目的 requirements.txt 或 pyproject.toml 以允许更高的 protobuf 版本。","https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F6718",{"id":150,"question_zh":151,"answer_zh":152,"source_url":153},43882,"在 Windows 上同时使用 PyTorch 和 ONNX 时，程序崩溃或导出失败，是否与导入顺序有关？","是的，这是一个已知的兼容性问题，特别是在 onnx >= 1.8.0 且使用静态链接 MSVC 运行时构建的版本中。\n临时解决方案：\n1. 调整代码中的导入顺序，务必在导入 torch 之前先导入 onnx。\n   例如：\n   import onnx\n   import torch\n   \n根本原因与 pybind11 处理异常及 protobuf 的静态\u002F动态链接冲突有关。长期解决方案是等待或编译使用动态 MSVC 运行时构建的 ONNX 包（ONNX 1.11+ 已逐步修复此问题）。","https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F3493",{"id":155,"question_zh":156,"answer_zh":157,"source_url":158},43883,"如何在 C\u002FC++ 中读取和操作 ONNX 模型？有类似 Python 的便捷 API 吗？","ONNX 官方提供了 C++ API，但相比 Python 接口较为底层。主要的操作方式是通过 ONNX Runtime 的 C API 或直接使用 protobuf 生成的 C++ 代码解析模型文件。\n推荐资源：\n1. 参考 ONNX Runtime 的 C API 文档（通常在 onnxruntime 仓库的 docs\u002FC_API.md）。\n2. 使用 protoc 编译器根据 onnx.proto 文件生成 C++ 代码来直接读取模型结构。\n注意：C++ 接口主要用于推理和底层操作，若需进行复杂的模型编辑或可视化，建议通过 Python 绑定处理或使用专门的 C++ 工具链。","https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F418",{"id":160,"question_zh":161,"answer_zh":162,"source_url":138},43884,"在 Windows ARM64 设备上安装 ONNX 后找不到 msvcp140.dll 导致运行失败，如何处理？","在 Windows ARM64 设备（如 Surface Laptop 7）上，标准的 VC++  redistributable 安装包有时不会将 msvcp140.dll 放置在被期望的系统目录中。\n解决方法：\n1. 搜索系统中存在的 dll 文件：dir \u002Fs \u002Fb msvcp140.dll\n2. 常见位置包括 C:\\Windows\\SysWOW64\\ 或 C:\\Windows\\System32\\Microsoft-Edge-WebView\\ 等子目录。\n3. 如果确认文件存在但不在系统路径，可以尝试将该文件复制到应用程序运行的目录，或者重新安装针对 ARM64 架构优化的最新 VC++ 运行库。\n4. 也可以尝试通过 winget 安装：winget install Microsoft.VCRedist.2015+.x64 (注意选择正确的架构版本)。",[164,169,174,179,184,189,194,199,204,209,214,219,224,229,234,239,244,249,254,259],{"id":165,"version":166,"summary_zh":167,"released_at":168},351320,"v1.21.0","ONNX v1.21.0 现已发布，带来了令人兴奋的新功能！我们衷心感谢所有为本次发布做出贡献的开发者！\n请访问 [onnx.ai](https:\u002F\u002Fonnx.ai\u002F)，了解更多关于 ONNX 及其相关项目的信息。\n\n## 变更内容\n### 重大变更与弃用\n* 移除已废弃的模型中心集成，由 @andife 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7678 中完成。\n### 规范与算子\n* 撤销 opset 25 版本号的提升，由 @titaiwangms 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7441 中完成。\n* 为 ONNX 添加 2 位支持，由 @vraspar 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7446 中完成。\n* 将 opset 版本提升至 26，由 @titaiwangms 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7469 中完成。\n* 移除对节点确定性属性的强制要求，由 @titaiwangms 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7473 中完成。\n* 导出函数，由 @cyyever 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7505 中完成。\n* [重新提交] 累积乘法算子，由 @titaiwangms 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7400 中完成。\n* 修复整数除法语义，改为使用截断除法，由 @MagellaX 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7585 中完成。\n* 明确当 inverse=True 且 onesided=True 时 DFT 的行为（irfft），由 @simonbyrne 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7574 中完成。\n* 规定 Lp 归一化中 0\u002F0 的情况，由 @titaiwangms 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7618 中完成。\n* 明确 NonMaxSuppression IoU 阈值采用严格比较，由 @Copilot 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7619 中完成。\n* 将模型域的要求由“必须”放宽为“建议”，由 @Copilot 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7668 中完成。\n* 添加 BitCast 算子，用于按位类型重新解释，由 @Copilot 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7527 中完成。\n### 参考实现\n* 修复 CenterCropPad：比较目标形状与维度，而非轴索引，由 @Copilot 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7522 中完成。\n* 对 sigmoid 算子实现进行向量化优化，由 @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7556 中完成。\n* 修复 reference_evaluator 在 verbose=3 时的冗余日志错误，由 @rothej 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7597 中完成。\n* 清理参考评估器测试，由 @VedantMadane 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7684 中完成。\n### 工具与实用程序\n* 为 test_executable_permission_check 添加版本检查，由 @andife 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7511 中完成。\n* 修复版本转换适配器中的索引越界问题，由 @rothej 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7493 中完成。\n* 修复形状推断中的默认属性值，由 @matteosal 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7602 中完成。\n* 修复 Softmax 系列算子对标量输入的形状推断问题，由 @MagellaX 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7625 中完成。\n* 修复并合并模型的 value info，由 @MagellaX 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7608 中完成。\n* 为 opset 15、16、17、18 添加版本转换适配器，由 @alexkvitkevich 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7677 中完成。\n* shape_inference：在 ParseData 中验证标量张量的预期大小，由 @skottmckay 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7675 中完成。\n### 构建、CI 与测试\n* 强制 PR 中添加标签，由 @titaiwangms 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7442 中完成。\n* 修复复用问题，由 @a","2026-03-27T20:35:27",{"id":170,"version":171,"summary_zh":172,"released_at":173},351321,"v1.20.1","\u003C!-- 发布说明由 .github\u002Frelease.yml 中 rel-1.20.1 配置生成 -->\n\n> [!注意]\n> 此补丁版本包含对 ONNX 构建的重要错误修复。\n\n## 变更内容\n\n* 导出函数（https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7505）\n* 撤销 ONNX CMake 的更改（https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7515）\n* 导出 PyTorch 中使用的函数（https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7524）\n* 导出 Protocol Buffers 符号（https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7525）\n\n### 提交记录\n* 由 @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7561 中将版本号提升至 1.20.1rc1\n* 由 @cyyever 和 @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7560 中将相关提交挑选并合并到 1.20.1 分支\n* 由 @titaiwangms 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7567 中移除 1.20.1 版本的 rc 标记\n\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fcompare\u002Fv1.20.0...v1.20.1","2026-01-09T23:21:12",{"id":175,"version":176,"summary_zh":177,"released_at":178},351322,"v1.20.0","ONNX v1.20.0 现已发布，带来了令人兴奋的新功能！我们衷心感谢所有为本次发布做出贡献的开发者！\n请访问 [onnx.ai](https:\u002F\u002Fonnx.ai\u002F) 了解更多关于 ONNX 及其相关项目的信息。\n\n## 更新的算子列表：\n\n[Cast](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Cast.html#cast-25)、[CastLike](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__CastLike.html#castlike-25)、[Constant](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Constant.html#constant-25)、[ConstantOfShape](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__ConstantOfShape.html#constantofshape-25)、[DequantizeLinear](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__DequantizeLinear.html#dequantizelinear-25)、[Flatten](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Flatten.html#flatten-25)、[Identity](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Identity.html#identity-25)、[If](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__If.html#if-25)、[Loop](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Loop.html#loop-25)、[Pad](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Pad.html#pad-25)、[QuantizeLinear](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__QuantizeLinear.html#quantizelinear-25)、[Reshape](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Reshape.html#reshape-25)、[Scan](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Scan.html#scan-25)、[Shape](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Shape.html#shape-25)、[Size](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Size.html#size-25)、[Squeeze](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Squeeze.html#squeeze-25)、[Transpose](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Transpose.html#transpose-25)、[Unsqueeze](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Unsqueeze.html#unsqueeze-25)\n\n## 主要更新：\n\n- 通过 Python 的稳定 ABI 支持 Python 3.14。([#7276](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7276))\n- Opset 25\n- 2 位数据类型支持 ([#7446](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7446))\n- 在算子模式中新增“节点确定性”属性 ([#7176](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7176))\n\n### 破坏性变更与弃用\n* 将 manylinux_2014 更新为 manylinux_2_28 ([#7151](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7151))\n* 更新注意力 GQA，使其在重复 KV 时使用重复插值法 ([#7274](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7274))\n* 将所需的 Python 版本更新至 3.10，并进行相关修复 ([#7220](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7220))\n* 移除 Python 3.9 的 wheel 构建 ([#7217](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7217))\n* 移除已弃用的方法 ([#7214](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7214))\n\n### 规范与算子\n* 为 ONNX 添加 2 位支持 ([#7446](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7446))\n* 移除对节点确定性属性的强制要求 ([#7473](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7473))\n* 修复 Softmax 算子对空输入的处理问题 ([#7206](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7206))\n* 修复 OneHotEncoder 因缺少输入形状验证而导致的段错误 ([#7302](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7302))\n* 修复注意力后端测试中 Range 输入的维度问题 ([#7300](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7300))\n* 修复注意力算子中 Range 输入的秩问题","2025-12-01T17:14:38",{"id":180,"version":181,"summary_zh":182,"released_at":183},351323,"v1.19.1","\u003C!-- 发布说明由 .github\u002Frelease.yml 中 rel-1.19.1 配置生成 -->\n\n> [!注意]\n> 此补丁版本包含对 Group Query Attention 模式下 Attention-23\u002F24 函数定义以及 RotaryEmbedding-23 参考实现的重要错误修复。\n\n## 所有变更\n\n* 避免不必要的 proto 文件重新生成 (#7253) 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7306 中\n* 要求 ml_dtypes>=0.5.0 (#7254) 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7307 中\n* 筛选并合并四个注意力相关的 Pull Request 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7315 中\n* 更新 rotary_embedding 参考实现和测试 (#7304, #7316) 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7313 中\n* 重写部分 proto 类的 `__repr__` 方法 (#7259) 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7314 中\n* 添加对 rc 候选版本的检查（更新 create_release.yml）(#7261) 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7323 中\n* 实现 Model\u002FGraph\u002FFunction 的 repr 方法 (#7320) 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7325 中\n","2025-10-10T02:06:42",{"id":185,"version":186,"summary_zh":187,"released_at":188},351324,"v1.19.0","\u003C!-- 发行说明由 .github\u002Frelease.yml 中 rel-1.19.0 配置生成 -->\nONNX v1.19.0 现已发布，带来了令人振奋的新特性！我们衷心感谢所有为本次发布做出贡献的开发者！\n请访问 [onnx.ai](https:\u002F\u002Fonnx.ai\u002F) 了解更多关于 ONNX 及其相关项目的信息。\n\n# 主要更新\n## IR 版本 12\n  * 新增 FLOAT8E8M0 类型\n## ai.onnx Opset 24\n  * 新增 Swish 操作\n  * 新增 TensorScatter 操作，并更新了 Attention 操作以支持就地 KV 缓存更新\n  * 为 QuantizeLinear、DequantizeLinear、Cast、CastLike、Constant、ConstantOfShape、Identity、Reshape、Shape、Size、If、Loop、Scan、Flatten、Pad、Squeeze、Unsqueeze 和 Transpose 启用 FLOAT8E8M0 数据类型。\n  * 为 TopK 和 SplitToSequence 启用 BF16 数据类型。\n## 其他\n  * 增加了对 ml-dtypes 的依赖\n  * `BUILD_ONNX_PYTHON` 符号已被弃用（将在 1.20 版本中移除）。请改用 `ONNX_BUILD_PYTHON`。\n\n## 变更内容\n### 中断性变更与弃用\n* @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6803 中弃用 helper 中的 printable_graph 函数\n* @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6914 中移除已弃用的映射常量\n* @cbourjau 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7083 中移除 re2 依赖\n* @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7089 中将 ml_dtypes 应用于所有地方\n### 规范与算子\n* @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6888 中明确 [un]Squeeze 的 `axes` 输入应为 1D 张量\n* @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6955 中明确不允许变量遮蔽\n* @cbourjau 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6973 中澄清 Mod 算子\n* @yuanyao-nv 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6984 中修正规范中关于 Attention 缩放因子的拼写错误\n* @robertknight 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7032 中澄清 Dropout 算子 `ratio` 输入的默认值\n* @robertknight 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7031 中修正 EyeLike 算子 `dtype` 属性的文档\n* @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7085 中更新 Cast 算子规范中的 float8 表格\n* @Copilot 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7056 中在 IR.md 中记录多设备配置协议规范\n* @yuanyao-nv 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7030 中新增 FLOAT8E8M0 数据类型\n* @yuanyao-nv 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7120 中为 Q\u002FDQ 及其他算子启用 float8e8m0\n* @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7130 中更新 Cast 和 CastLike 中 E4M3FNUZ\u002FE5M2FNUZ 的饱和行为\n* @Copilot 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7136 中修复 ELU 和 Softplus 算子，使其支持任意形状的张量\n* @Copilot 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7132 中修复 Shape 算子规范：修正范围边界并记录 start > end 的行为\n* @xadupre 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7142 中修复 Attention 3D、参考实现及 C++ 扩展\n* @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F7136 中修复 RMS 归一化函数定义","2025-08-26T21:48:02",{"id":190,"version":191,"summary_zh":192,"released_at":193},351325,"v1.18.0","\u003C!-- 发布说明由 justinchu\u002Frel-118-notes 仓库中 .github\u002Frelease.yml 文件中的配置生成 -->\n\nONNX v1.18.0 现已发布，带来了令人振奋的新特性！我们衷心感谢所有为本次发布做出贡献的开发者和社区成员！  \n请访问 [onnx.ai](https:\u002F\u002Fonnx.ai\u002F)，了解更多关于 ONNX 及其相关项目的信息。\n\n# 重要更新\n## ai.onnx Opset 23\n[Attention](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Attention.html#attention-23)、[Cast](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Cast.html#cast-23)、[CastLike](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__CastLike.html#castlike-23)、[Constant](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Constant.html#constant-23)、[ConstantOfShape](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__ConstantOfShape.html#constantofshape-23)、[DequantizeLinear](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__DequantizeLinear.html#dequantizelinear-23)、[Flatten](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Flatten.html#flatten-23)、[Identity](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Identity.html#identity-23)、[If](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__If.html#if-23)、[Loop](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Loop.html#loop-23)、[Pad](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Pad.html#pad-23)、[QuantizeLinear](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__QuantizeLinear.html#quantizelinear-23)、[RMSNormalization](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__RMSNormalization.html#rmsnormalization-23)、[Reshape](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Reshape.html#reshape-23)、[RotaryEmbedding](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__RotaryEmbedding.html#rotaryembedding-23)、[Scan](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Scan.html#scan-23)、[Shape](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Shape.html#shape-23)、[Size](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Size.html#size-23)、[Squeeze](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Squeeze.html#squeeze-23)、[Transpose](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Transpose.html#transpose-23)、[Unsqueeze](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Unsqueeze.html#unsqueeze-23)\n\n## IR 版本 11\n  * 新增 FLOAT4E2M1 数据类型及多设备配置支持\n  * 放宽了命名规则要求（https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6652）\n\n## Python 支持\n  * 完全支持 Python 3.13\n  * 实验性支持 Python 3.13t（Windows 和 Mac 平台）\n  * 移除对 Python 3.8 的支持\n  * 提供适用于 Windows Arm64 架构的预编译轮子包\n\n## 构建改进\n  * 将最低 protobuf 版本升级至 v25.1\n  * CMake 中新增 ONNX_BUILD_CUSTOM_PROTOBUF 选项（https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6495）\n\n## 其他变更\n### 破坏性变更与弃用\n* 在调用 onnx.save 时，若存在外部文件则移除或抛出异常——由 @tonypottera24 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6497 中实现\n* 移除 Python 3.8 的 CI 流水线（因 Python 3.8 已停止维护）——由 @cyyever 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6434 中完成\n* 弃用所有类型转换函数——由 @justinchuby 在 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F6639 中提出，计划于 1.20 版本中正式移除\n* GroupNormalization-18 现已弃用，并被 GroupNormalization-23 取代，原因是定义存在错误 ([","2025-05-12T23:07:37",{"id":195,"version":196,"summary_zh":197,"released_at":198},351326,"v1.17.0","ONNX 1.17.0 现已发布，带来了令人振奋的新特性！我们衷心感谢所有为本次发布做出贡献的开发者和社区成员！\n请访问 [onnx.ai](https:\u002F\u002Fonnx.ai\u002F)，了解更多关于 ONNX 及其相关项目的信息。\n\n# 重要更新\n## ai.onnx Opset 22\n* 更新以支持 bfloat16：\n  * [Acos](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Acos.html#acos-22)、[Acosh](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Acosh.html#acosh-22)、[Asin](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Asin.html#asin-22)、[Asinh](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Asinh.html#asinh-22)、[Atan](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Atan.html#atan-22)、[Atanh](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Atanh.html#atanh-22)、[AveragePool](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__AveragePool.html#averagepool-22)、[Bernoulli](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Bernoulli.html#bernoulli-22)、[Conv](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Conv.html#conv-22)、[ConvTranspose](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__ConvTranspose.html#convtranspose-22)、[Cos](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Cos.html#cos-22)、[Cosh](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Cosh.html#cosh-22)、[DeformConv](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__DeformConv.html#deformconv-22)、[Det](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Det.html#det-22)、[Dropout](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Dropout.html#dropout-22)、[Elu](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Elu.html#elu-22)、[EyeLike](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__EyeLike.html#eyelike-22)、[GRU](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__GRU.html#gru-22)、[GlobalAveragePool](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__GlobalAveragePool.html#globalaveragepool-22)、[GlobalLpPool](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__GlobalLpPool.html#globallppool-22)、[GlobalMaxPool](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__GlobalMaxPool.html#globalmaxpool-22)、[GridSample](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__GridSample.html#gridsample-22)、[HardSigmoid](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__HardSigmoid.html#hardsigmoid-22)、[HardSwish](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__HardSwish.html#hardswish-22)、[InstanceNormalization](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__InstanceNormalization.html#instancenormalization-22)、[LSTM](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__LSTM.html#lstm-22)、[LpNormalization](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__LpNormalization.html#lpnormalization-22)、[LpPool](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__LpPool.html#lppool-22)、[MaxPool](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__MaxPool.html#maxpool-22)、[MaxRoiPool](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__MaxRoiPool.html#maxroipool-22)、[MaxUnpool](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__MaxUnpool.html#maxunpool-22)、[Mish](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Mish.html#mish-22)、[Multinomial](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Multinomial.html#multinomial-22)、[NegativeLogLikelihoodLoss](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__NegativeLogLikelihoodLoss.html#negativeloglikelihoodloss-22)、[RNN](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__RNN.html#rnn-22)、[RandomNormal](ht","2024-10-01T17:57:22",{"id":200,"version":201,"summary_zh":202,"released_at":203},351327,"v1.16.2","ONNX 1.16.2 是基于 1.16.1 的补丁版本。\n\n# 错误修复\n* 降低 tarball 目录遍历风险 #6164\n* 重构安全解压方法 #6222\n* 在 Gemm 算子集 6 中添加维度检查 #6217\n* 更新损坏的 URL #6255\n\n请访问 [onnx.ai](https:\u002F\u002Fonnx.ai) 以了解更多关于 ONNX 及相关项目的信息。","2024-08-01T13:15:32",{"id":205,"version":206,"summary_zh":207,"released_at":208},351328,"v1.16.1","ONNX 1.16.1 是基于 1.16.0 的补丁版本。\n\n# 错误修复\n* 防止在使用 GCC 8 编译后导入时发生崩溃 #6048\n* 为 DequantizeLinear 添加缺失的形状推断检查 #6080\n* 修复量化\u002F反量化的 ONNX 后端测试中的输入名称问题 #6122\n* 修复遗漏的形状推断代码 #6049\n\n请访问 [onnx.ai](https:\u002F\u002Fonnx.ai) 以了解更多关于 ONNX 及其相关项目的信息。","2024-05-23T17:50:05",{"id":210,"version":211,"summary_zh":212,"released_at":213},351329,"v1.16.0","ONNX v1.16.0 现已发布，带来了令人兴奋的新功能！我们衷心感谢所有为本次发布做出贡献的开发者！请访问 [onnx.ai](https:\u002F\u002Fonnx.ai) 了解更多关于 ONNX 及其相关项目的信息。\n\n# 重要更新\n## ai.onnx Opset 21\n* 更新以支持 int4 和 uint4：\n    * [Cast](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Cast.html#cast-21)、[CastLike](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__CastLike.html#CastLike-21)、[Constant](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Constant.html#Constant-21)、[ConstantOfShape](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__ConstantOfShape.html#ConstantOfShape-21)、[Identity](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Identity.html#Identity-21)、[If](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__If.html#If-21)、[Loop](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Loop.html#Loop-21)、[Reshape](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Reshape.html#Reshape-21)、[Scan](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Scan.html#Scan-21)、[Shape](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Shape.html#Shape-21)、[Size](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Size.html#Size-21)\n* 更新以支持 float8e4m3fnuz、float8e5m2、float8e5m2fnuz、int4 和 uint4：\n    * [Flatten](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Flatten.html#Flatten-21)、[Pad](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Pad.html#Pad-21)、[Squeeze](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Squeeze.html#Squeeze-21)、[Transpose](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Transpose.html#Transpose-21)、[Unsqueeze](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__Unsqueeze.html#Unsqueeze-21)\n* 支持分块量化。支持 int4、uint4、int16 和 uint16：\n    * [DequantizeLinear](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__DequantizeLinear.html#DequantizeLinear-21)、[QuantizeLinear](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__QuantizeLinear.html#QuantizeLinear-21)\n* 支持 bfloat16 和 float16 缩放。支持 float8e4m3fn、float8e4m3fnuz、float8e5m2、float8e5m2fnuz 量化张量：\n    * [QLinearMatMul](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__QLinearMatMul.html#QLinearMatMul-21)\n* 为 [GroupNormalization](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx__GroupNormalization.html#GroupNormalization-21) 添加 `stash_type` 属性，并将 `scale` 和 `bias` 的输入形状由 (G) 更改为 (C)\n\n## ai.onnx.ml Opset 4\n* 新增运算符 [TreeEnsemble](https:\u002F\u002Fonnx.ai\u002Fonnx\u002Foperators\u002Fonnx_aionnxml_TreeEnsemble.html#treeensemble-5-ai-onnx-ml)\n\n## IR 版本 10\n* 增加对 UINT4、INT4 类型的支持\n* GraphProto、FunctionProto、NodeProto、TensorProto 新增 `metadata_props` 字段\n* FunctionProto 新增 `value_info` 字段\n* FunctionProto 和 NodeProto 新增 `overload` 字段，以支持重载函数\n\n## Python 方面的变更\n* 支持通过 Python 接口注册自定义 OpSchema\n* 支持 Python 3.12\n\n## 安全更新\n* 修复路径清理绕过漏洞，该漏洞可能导致任意读取（CVE-2024-27318）\n* 修复因断言语句中缺少字符串终止符而导致的越界读取问题（CVE-2024-27319）\n\n# 废弃通知\n* 已废弃使用 C++14 编译 ONNX","2024-03-25T15:40:18",{"id":215,"version":216,"summary_zh":217,"released_at":218},351330,"v1.15.0","ONNX v1.15.0 is now available with exciting new features! We would like to thank everyone who contributed to this release! Please visit [onnx.ai](https:\u002F\u002Fonnx.ai) to learn more about ONNX and associated projects.\r\n\r\n# Key Updates\r\n- Added new operators: [ImageDecoder](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#ImageDecoder)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5294 [RegexFullMatch](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#RegexFullMatch)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5401 [StringConcat](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#StringConcat)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F5350 [StringSplit](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#StringSplit)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5371 [AffineGrid](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#AffineGrid)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F5225 [Gelu](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#Gelu)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F5277\r\n\r\n- Updated existing operators: [ConstantOfShape](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#ConstantOfShape-20)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5390 [GridSample](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#GridSample-20)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5010 [ReduceMax](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#ReduceMax-20)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5539 [ReduceMin](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#ReduceMin-20)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5539 [IsNan](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#IsNan-20)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5583 [IsInf](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#IsInf-20)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5583 [DFT](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#DFT-20)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5514 [LabelEncoder](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog-ml.md#ai.onnx.ml.LabelEncoder-4)https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5453\r\n\r\n- New features, bug fixes, and document updates\r\n \r\n## ai.onnx opset version increased to 20 with following changes:\r\n- New Operators (ai.onnx): \r\n\t- [ImageDecoder](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#ImageDecoder) a new ImageDecoder operator to be used in preprocessing models\r\n\t- [RegexFullMatch](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#RegexFullMatch) a new operator for regex matching that is commonly used in feature preprocessing\r\n\t- [StringConcat](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#StringConcat) takes two string tensors as input and returns the elementwise concatenation of the strings in each tensor\r\n\t- [StringSplit](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#StringSplit) takes a string tensor as input and splits each element based on a delimiter attribute and a maxsplit attribute\r\n\t- [AffineGrid](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#AffineGrid) Generates a 2D or 3D flow field (sampling grid), given a batch of affine matrices theta\r\n\t- [Gelu](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#Gelu) applies gaussian error linear unit function or its approximation to input\r\n\r\n- Operator Updates (ai.onnx):\r\n\t- [ConstantOfShape](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#ConstantOfShape) extends supported data types\r\n\t- [GridSample](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#GridSample) extends to ND data\r\n\t- [ReduceMax](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#ReduceMax) adds support for boolean\r\n\t- [ReduceMin](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#ReduceMin) adds support for boolean\r\n\t- [IsNan](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#IsNan) adds support of float8 types\r\n\t- [IsInf](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#IsInf) adds support of float8 types\r\n\t- [DFT](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#DFT) promote axis as input\r\n\r\n\r\n## ai.onnx.ml opset version increased to 4 with following changes:\r\n- Operator Updates (ai.onnx.ml):  \r\n\t- [LabelEncoder](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog-ml.md#ai.onnx.ml.LabelEncoder-4) adds keys_as_tensor and values_as_tensor attributes\r\n\r\n## New functionality:\r\n- Enable empty list of values as attribute [PR#5559](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5559)\r\n- Update diff bakend node tests for auto update doc [PR#5604](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5604)\r\n- Enable pylint checks with Ruff and remove pylint from lintrunner [PR#5589](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5589)\r\n- Getting onnx to treat `inf\u002F-inf` as float literals. [PR#5528](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5528)\r\n- Create the onnxtxt serialization format [PR#5524](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5524)\r\n- Support JSON as a serialization target [PR#5523](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5523)\r\n- Support for parsing and printing empty list value as attribute [PR#5516](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5516)\r\n- Add auto updat","2023-10-31T17:04:56",{"id":220,"version":221,"summary_zh":222,"released_at":223},351331,"v1.14.1","ONNX v1.14.1 is a patch release based on v1.14.1.\r\n\r\n# Bug fixes\r\n- Fix `shape` data propagation function to handle missing optional parameters #5219\r\n- Fix a couple of shape inference issues #5223\r\n- Extend function type inference to handle missing optional parameters #5169\r\n- Fix check_tensor to work with large models on Windows #5227\r\n- Fix check_tensor to work with large models on UNIX #5286","2023-08-25T21:00:03",{"id":225,"version":226,"summary_zh":227,"released_at":228},351332,"v1.14.0","ONNX v1.14.0 is now available with exciting new features! We would like to thank everyone who contributed to this release! Please visit [onnx.ai](https:\u002F\u002Fonnx.ai\u002F) to learn more about ONNX and associated projects.\r\n\r\n# Opset 19 is released\r\n## New operators\r\n[DeformConv](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#deformconv-19) added in https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4783\r\n\r\n## Operator extensions\r\n[Equal](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#equal-19) - Support for string data type added in https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4828\r\n[AveragePool](url) - New attribute ``dilations`` https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4790\r\n[Pad](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#pad-19) - Added new ``wrap`` to the ``mode`` attribute to support circular padding https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4793\r\n[Resize](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4862) - Added ``half_pixel_symmetric`` to the ``coordinate_transformation_mode`` attribute https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4862\r\n\r\n# IR updates (bump to 9)\r\n- Support attributes with default values: https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4911\r\n- Added 4 new 8-bit floating point data types: https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4805\r\n\r\n# Backend tests\r\nReplaced real models with light models in backend tests. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4861 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4960\r\n\r\n# Support Protobuf v21\r\nNow ONNX supports Protobuf v21: https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4956\r\n\r\n# Deprecation notice\r\n- Python 3.7 support will be deprecated due to EOL in next release: https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F5191\r\n- onnx-weekly package will be deprecated in TestPyPI. Please use them from PyPI instead: https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F4930\r\n- Properties in FormalParameter will be deprecated in future release. Please use newer properties name:  https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F5074\r\n- Variables from mapping.py will be deprecated and become private implementation details. Please use public functions to get corresponding types from helper.py instead: https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F4554\r\n\r\n# Installation notice\r\nYou can upgrade to the latest release using `pip install onnx --upgrade` or build from source following the README [instructions](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Ftree\u002Frel-1.14.0#build-onnx-from-source).\r\n\r\n# Contributors\r\nThanks to these individuals for their contributions in this release since last 1.13.0 release: @jcwchen, @andife, @gramalingam, @xadupre, @justinchuby, @liqunfu, @yuanyao-nv, @jbachurski, @p-wysocki, @prasanthpul, @jantonguirao, @take-cheeze, @smk2007, @AlexandreEichenberger, @snnn, @daquexian, @linkerzhang. ","2023-05-05T16:35:02",{"id":230,"version":231,"summary_zh":232,"released_at":233},351333,"v1.13.1","ONNX v1.13.1 is a patch release based on v1.13.0.\r\n\r\n# Bug fixes\r\n- Add missing f-string for DeprecatedWarningDict in mapping.py  #4707\r\n- Fix types deprecated in numpy==1.24 #4721 \r\n- Update URL for real models from ONNX Runtime #4865\r\n- Fix attribute substitution within subgraphs during function type\u002Fshape inference #4792\r\n- Handle variants of constant op in shape inference #4824\r\n- Fix parser bug in handling non-tensor types #4863\r\n- Fix function shape inference bug #4880\r\n\r\n# Announcement\r\n- Deprecate real model tests from onnx repo in next ONNX release #4885\r\n- Move onnx-weekly package from TestPyPI to PyPI and stop uploading onnx-weekly to TestPyPI after next ONNX release #4930","2023-02-22T18:47:36",{"id":235,"version":236,"summary_zh":237,"released_at":238},351334,"v1.13.0","ONNX v1.13.0 is now available with exciting new features! We would like to thank everyone who contributed to this release! Please visit [onnx.ai](https:\u002F\u002Fonnx.ai\u002F) to learn more about ONNX and associated projects.\r\n\r\n# New operators\r\n- [Col2Im](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#Col2Im-18) added in [#3948](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3948)\r\n- [BitwiseNot](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#bitwisenot-18) added in [#4497](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4497)\r\n- [BitwiseAnd](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#bitwiseand-18), [BitwiseOr](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#bitwiseor-18) and [BitwiseXor](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#bitwisexor-18) added in [#4496](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4496)\r\n\r\n# Operator extensions\r\n- [Resize](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#resize-18) - New attributes: `antialias`, `axes` and `keep_aspect_ratio_policy`, allow for both `scales` and `sizes` to be provided when one of them is an empty constant [#4126](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4126), [#4388](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4388)\r\n- [Pad](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#pad-18) - New attribute `axes` [#4190](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4190)\r\n- [OptionalHasElement](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#optionalhaselement-18) - New input types handling [#4326](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4326)\r\n- [OptionalHasElement](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#optionalhaselement-18) and [OptionalGetElement](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#optionalgetelement-18) - Accept tensor and sequence types [#4421](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4421)\r\n- [ScatterElement](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#scatterelements-18) and [ScatterND](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#scatternd-18) - Add `max` and `min` as supported reduction attributes [#4411](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4411)\r\n- [Split](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#split-18) - Add support for uneven tensor splitting and a new `num_outputs` attribute [#4481](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4481)\r\n- [LpPool](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#lppool-18) - New attributes: `ceil_mode` and `dilations` [#4534](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4534)\r\n\r\n# Function updates\r\n- [CenterCropPad](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#centercroppad-18) added in [#4190](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4190)\r\n- [mish](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#mish-18) added in [#4350](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4350)\r\n- [GroupNormalization](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#groupnormalization-18) added in [#4621](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4621)\r\n\r\n# Reference Python runtime\r\nReference Python runtime dependent on only Python and numpy has been added. [#4483](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4483)\r\n\r\n# Python 3.11 support\r\nONNX 1.13.0 supports Python 3.11. [#4490](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4490)\r\n\r\n# Apple Silicon support\r\nSupport for M1\u002FM2 ARM processors has been added. [#4642](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4642)\r\n\r\n# More\r\nONNX 1.13.0 also comes with numerous:\r\n- bugfixes\r\n- infrastructure improvements\r\n- CI improvements\r\n- documentation updates\r\n- security updates\r\n\r\nFor full details see [Logistics for ONNX Release 1.13.0](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fwiki\u002FLogistics-for-ONNX-Release-1.13.0).\r\n\r\n# Deprecation notice\r\n- `TENSOR_TYPE_TO_STORAGE_TENSOR_TYPE` has been deprecated [#4270](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4270)\r\n- ONNXIFI: ONNX Interface for Framework Integration has been deprecated [#4431](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F4431)\r\n\r\n# Installation\r\nYou can upgrade to the latest release using `pip install onnx --upgrade` or build from source following the README [instructions](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Ftree\u002Frel-1.13.0#build-onnx-from-source).\r\n\r\n# Contributors\r\nThanks to these individuals for their contributions in this release since last 1.12.0 release: @AnandKri, @cbourjau, @jcwchen, @gramalingam, @garymm, @GaetanLepage, @ilya-lavrenov, @jnovikov, @JackBoosY, @jbachurski, @tjich, @jantonguirao, @justinchuby, @natke, @philass, @prasanthpul, @p-wysocki, @SpaceIm, @stephenneuendorffer,@take-cheeze, @sechkova, @thiagocrepaldi, @xadupre, @mszhanyi, @yuanyao-nv, @andife, @daquexian, @kylesayrs, @liqunfu, @longlee0622, @HSQ79815, @williamberman, @YanBC\r\n\r\nThe list has been acquired with a [script](https:\u002F\u002Fgist.github.com\u002Fabock\u002F30fc4e8b1fd1e879e113597ff536aa35) written by [Aaron Bockover](https:\u002F\u002Fgithub.com\u002Fabock).","2022-12-12T14:38:36",{"id":240,"version":241,"summary_zh":242,"released_at":243},351335,"v1.12.0","ONNX v1.12.0 is now available with exciting new features! We would like to thank everyone who contributed to this release! Please visit [onnx.ai](https:\u002F\u002Fonnx.ai) to learn more about ONNX and associated projects.\r\n\r\n# Key Updates\r\n- Added new operators: [SequenceMap](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#SequenceMap), [LayerNormalization function operator](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#LayerNormalization), [DFT](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#DFT), [HannWindow](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#HannWindow), [HammingWindow](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#HammingWindow), [BlackmanWindow](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#BlackmanWindow), [MelWeightMatrix](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#MelWeightMatrix), [STFT](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#STFT)\r\n- Updated existing operators:  [Scan](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#Scan-16)\r\n- Miscellaneous shape inference enhancements. \r\n- Miscellaneous bugfixes and infrastructure improvements.\r\n- Miscellaneous documentation updates.\r\n- Wheel updates\r\n  - Add Python 3.10 and drop Python 3.6 support\r\n  - Now use newer manylinux2014_* instead of manylinux2010_* for Linux environments. Please check the support list [here](https:\u002F\u002Fgithub.com\u002Fpypa\u002Fmanylinux#manylinux).\r\n  - Drop support for x86 (32-bit) Linux due to low usage\r\n\r\n## ai.onnx opset version increased to 17 with following changes:\r\n- New operators (ai.onnx):\r\n        - LayerNormalization (#4076)\r\n        - SequenceMap (#3892)\r\n        - Signal Operators: DFT, HannWindow, HammingWindow, BlackmanWindow, MelWeightMatrix, STFT (#3741)\r\n- Operator Updates (ai.onnx):  \r\n        - [Scan] Remove unused type constraint I for newer Scan (opset 9+)(#4012)\r\n\r\n## Shape inference enhancements\r\n- Extend InferShapes to expose result of data propagation (#3879)\r\n- Update shape inference for constant of shape (#4141)\r\n- Catch missing input type in function shape inference (#4123)\r\n- Add shape inference for Expand using symbolic shape input (#3789)\r\n- Fix Expand shape inference: stop rank inference if the shape is symbolic (#4019)\r\n\r\n## Bug fixes and infrastructure improvements\r\n- Fix a bug in _get_initializer_tensors() (#4118)\r\n- Fix bug of resizeShapeInference for Resize13 (#4140)\r\n- Fix bug in SCE function body (#4038)\r\n- Use correct pytest types in backend (#3990) (#3994)\r\n- Checker should validate the node's inputs\u002Foutputs have names when its formal parameter is Variadic  (#3979)\r\n- Loose NumPy requirement to grant more flexibility (#4059)\r\n- Fix crash: Skip unused value_info for version_converter (#4079)\r\n- Use %d for integer in version_converter (#4182)\r\n- Extend parser to handle other types (#4136)\r\n\r\n## Documentation updates\r\n- Add documentation about functions to IR.md (#4180)\r\n- Clarify add new op documentation (#4150)\r\n- Clarify NonZero behavior for scalar input in spec (#4113)\r\n- Update shape inference documentation (#4163)\r\n- Fix a minor typo in operator Gather documentation (#4125)\r\n- Fix typo in CIPipelines.md (#4157)\r\n- Fix typo in slice doc (#4117)\r\n- Fix grammar in documents (#4094)\r\n- Clearer description of Slice (#3908)\r\n- Add OperatorSetId definition in docs (#4039)\r\n- Clean up protocol buffer definitions (#4201)\r\n- Change the wrong words of second layer **`input`** (#4044)\r\n- Clarify that op_type is case sensitive (#4096)\r\n\r\n# Installation\r\nYou can upgrade to the latest release using `pip install onnx --upgrade` or build from source following the README [instructions](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Ftree\u002Frel-1.10.0#build-onnx-from-source).\r\n\r\n# Notes \r\n- Beware of the protobuf version gap issue (building onnx with protobuf>=3.12 is not compatible with older protobuf)\r\n\r\n# Contributors\r\nThanks to these individuals for their contributions in this release since last 1.11.0 release. (Contributor list obtained with: https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fgraphs\u002Fcontributors?from=2022-02-08&to=2022-05-24&type=c): @jcwchen, @gramalingam, @xuzijian629, @garymm, @diyessi, @liqunfu, @jantonguirao, @daquexian, @fdwr, @andife, @wschin, @xadupre, @xkszltl, @snnn\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n","2022-06-18T02:58:18",{"id":245,"version":246,"summary_zh":247,"released_at":248},351336,"v1.11.0","ONNX v1.11.0 is now available with exciting new features! We would like to thank everyone who contributed to this release! Please visit [onnx.ai](https:\u002F\u002Fonnx.ai) to learn more about ONNX and associated projects.\r\n\r\n# Key Updates\r\n- Added new operators: [GridSample](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#gridsample) https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3557)\r\n- Updated existing operators: [Identity](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#identity-16), [If](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#if-16), [LeakyRelu](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#leakyrelu-16), [Loop](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#loop-16), [PRelu](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#prelu-16), [RoiAlign](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#roialign-16), [Scan](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#scan-16), [ScatterElements](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#ScatterElements-16), [ScatterND](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#ScatterND-16), [Where](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#Where-16), [GreaterOrEqual](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#GreaterOrEqual-16), [LessOrEqual](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#LessOrEqual-16), [TreeEnsembleClassifier](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog-ml.md#aionnxmltreeensembleclassifier-3), [TreeEnsembleRegressor](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog-ml.md#aionnxmltreeensembleregressor-3)\r\n- Added Model Hub for users to get started with state-of-the-art pre-trained ONNX models from the ONNX Model Zoo or for researchers and model developers to share models. [#3712](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3712)\r\n- Added a set of compose utilities to create a combined model that includes both preprocessing and inferencing networks [#3820](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3820).\r\n- Added FunctionBuilder utility class to help construct function ops [#3882](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3882).\r\n- Miscellaneous shape inference enhancements. \r\n- Miscellaneous bugfixes and infrastructure improvements.\r\n- Miscellaneous documentation updates.\r\n\r\n## ai.onnx opset version increased to 16 with following changes:\r\n- New Operators (ai.onnx): \r\n\t- [GridSample](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FOperators.md#gridsample)\r\n- Operator Updates (ai.onnx):  \r\n\t- [Identity](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#identity-16), add optional type support. \r\n\t- [If](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#if-16), add optional data type support for output. \r\n\t- [LeakyRelu](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#leakyrelu-16), add bfloat16 type support. \r\n\t- [Loop](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#loop-16), add optional data type support for initial value and output.\r\n\t- [PRelu](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#prelu-16), add bfloat16 type support.\r\n\t- [RoiAlign](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#roialign-16), add an attribute coordinate_transformation_mode, correct the default behavior.\r\n\t- [Scan](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#scan-16), add bfloat16 type support for output.\r\n\t- [ScatterElements](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#ScatterElements-16), add reduction attribute.\r\n\t- [ScatterND](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#ScatterND-16), add reduction attribute.\r\n\t- [Where](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#Where-16), extend Where op to permit bfloat16 types.\r\n\t- [GreaterOrEqual](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#GreaterOrEqual-16), add bfloat16 type support.\r\n\t- [LessOrEqual](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog.md#LessOrEqual-16), add bfloat16 type support.\r\n\r\n## ai.onnx.ml opset version increased to 3 with following changes:\r\n- [TreeEnsembleClassifier](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog-ml.md#aionnxmltreeensembleclassifier-3), support tensor as attributes. \r\n- [TreeEnsembleRegressor](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmain\u002Fdocs\u002FChangelog-ml.md#aionnxmltreeensembleregressor-3), support tensor as attributes. \r\n\r\n## New functionality:\r\n- A new Model Hub for users to get started with state-of-the-art pre-trained ONNX models from the ONNX Model Zoo or for researchers and model developers to share models. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3712\r\n- Add compose utility to help with creating and combining models out of several graphs. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3820\r\n- Add FunctionBuilder utility class to help construct function ops. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3882\r\n\r\n## Shape inference enhancements\r\n- Extend optional type inference. [#3756](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3756)\r\n- Make shape inference handle MapProto. [#3772](https:\u002F\u002Fgith","2022-02-17T21:58:59",{"id":250,"version":251,"summary_zh":252,"released_at":253},351337,"v1.10.2","ONNX v1.10.2 is a patch release based on v1.10.1.\r\n\r\nBug fixes:\r\n\r\n* Fix compilation error on older compilers ([#3683](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3683))\r\n* Stricter check for Shape's input: check input type ([#3757](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3757))","2021-10-26T17:29:47",{"id":255,"version":256,"summary_zh":257,"released_at":258},351338,"v1.10.1","This release is a patch release based on v1.10.0.\r\n\r\nBug fix:\r\n- Include requirements.txt in source distribution https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3623","2021-08-02T20:22:19",{"id":260,"version":261,"summary_zh":262,"released_at":263},351339,"v1.10.0","ONNX v1.10.0 is now available with exciting new features! We would like to thank everyone who contributed to this release! Please visit [onnx.ai](https:\u002F\u002Fonnx.ai) to learn more about ONNX and associated projects.\r\n\r\n# Key Updates\r\n- Added new `Optional` and `SparseTensor` types. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3407 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3398\r\n- Added model local functions to ModelProto. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3532\r\n- Shape inference enhancements for `Reshape`, `Squeeze`, `NonZero`, `DynamicQuantizeLinear`.\r\n- Introduce symbolic shape inference support. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F3506\r\n- New version converter tests. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3344\r\n- Add aarch64 wheel build support. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3414\r\n- Update ONNX IR version to 8 and opset version to 15. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3587\r\n\r\n## IR Updates\r\n- Added two new types to ONNX type system. `Optional` and `SparseTensor` https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3407 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3398\r\n- Extend model proto to include model local functions. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3532\r\n\r\n## Opset version 15\r\n- New Function Operators:\r\n\t- [Bernoulli](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FOperators.md#Bernoulli) https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3431\r\n\t- [CastLike](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FChangelog.md#CastLike-15) https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3558\r\n- New Operators: \r\n\t- [Optional](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FChangelog.md#optional-15)\r\n\t- [OptionalGetElement](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FChangelog.md#optionalgetelement-15)\r\n\t- [OptionalHasElement](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FChangelog.md#optionalhaselement-15)\r\n- Operator Updates:  \r\n\t- Add additional type constraints in [BatchNormalization](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FChangelog.md#BatchNormalization-15). https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3545\r\n\t- Add `bfloat16` support for [Pow](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FChangelog.md#Pow-15). https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3412  \r\n\t- Extend [Shape](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fblob\u002Fmaster\u002Fdocs\u002FChangelog.md#Shape-15) to return a slice using optional attributes `start`,`end`. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3580  \r\n\r\n## API\r\n- Symbolic shape inference support. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fissues\u002F3506\r\n\t- Symbol generation https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3518\r\n\t- Data propagation https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3551 https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3593\r\n- Shape inference enhancements\r\n\t- Add shape inference for `NonZero`. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3364\r\n\t- Add shape inference for `Dynamic QuantizeLinear`. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3539\r\n\t- Update `Reshape` shape inference. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3592\r\n\t- Fix shape inference for `Squeeze`. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3516\r\n        - Fix shape inference for `Squeeze` without axes. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3465\r\n- Expose model parser API in Python (`onnx.parser`). https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3540\r\n- Extend model proto to include model local functions.\r\n\r\n\r\n## Infrastructure\r\n- Update protobuf version to 3.16. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3571\r\n- Add README contents to package description. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3376\r\n- Add requirements.txt to onnx repo. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3448\r\n- Add aarch64 wheel build support. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3414\r\n- Version converter support for recursion into subgraphs. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3474\r\n- Update ONNX examples to python3. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3450\r\n\r\n## Bug fixes\r\n- Spec clarification for `MatMulInteger` and `QLinearMatMul`. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3585\r\n- Extend `strict_model` for ONNX checker. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3348\r\n- Always set the output of `Shape` to be rank-1. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3394\r\n- `BatchNormalization` outputs updated for training mode. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3379\r\n- Bugfix for proto utils and update checker error messages. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3373\r\n- Fix compilation warnings. https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Fpull\u002F3616\r\n\r\n# Installation\r\nYou can upgrade to the latest release using `pip install onnx --upgrade` or build from source following the README [instructions](https:\u002F\u002Fgithub.com\u002Fonnx\u002Fonnx\u002Ftree\u002Frel-1.10.0#build-onnx-from-source).\r\n\r\n# Notes \r\n- Beware of the protobuf version gap issue (building onnx with protobuf>=3.12 is not compatible with older protobuf)\r\n\r\n# Contributors\r\nThanks to these individuals for their contributions in this release:\r\n@jcwchen, @askhade, @gramalingam, @neginraoof, @matteosal, @postrational, @garymm, @yuslepukhin, @fdwr, @jackwish, @manbearian, @etusien, @impactaky, @rajeevsrao, @prasanthpul, @take-cheeze, @chudegao, @mindest, @yufenglee, @annajung, @hwangdeyu, @calvinmccarter-at-lightmatter, @ashbhandare, @xuzijian629, @IceTDrinker, @mrry","2021-07-31T00:56:50"]