[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-sony--nnabla":3,"tool-sony--nnabla":61},[4,18,26,36,44,53],{"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",157379,2,"2026-04-15T23:32:42",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":32,"last_commit_at":42,"category_tags":43,"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。",108322,"2026-04-10T11:39:34",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":32,"last_commit_at":50,"category_tags":51,"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",[52,13,15,14],"插件",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":32,"last_commit_at":59,"category_tags":60,"status":17},4721,"markitdown","microsoft\u002Fmarkitdown","MarkItDown 是一款由微软 AutoGen 团队打造的轻量级 Python 工具，专为将各类文件高效转换为 Markdown 格式而设计。它支持 PDF、Word、Excel、PPT、图片（含 OCR）、音频（含语音转录）、HTML 乃至 YouTube 链接等多种格式的解析，能够精准提取文档中的标题、列表、表格和链接等关键结构信息。\n\n在人工智能应用日益普及的今天，大语言模型（LLM）虽擅长处理文本，却难以直接读取复杂的二进制办公文档。MarkItDown 恰好解决了这一痛点，它将非结构化或半结构化的文件转化为模型“原生理解”且 Token 效率极高的 Markdown 格式，成为连接本地文件与 AI 分析 pipeline 的理想桥梁。此外，它还提供了 MCP（模型上下文协议）服务器，可无缝集成到 Claude Desktop 等 LLM 应用中。\n\n这款工具特别适合开发者、数据科学家及 AI 研究人员使用，尤其是那些需要构建文档检索增强生成（RAG）系统、进行批量文本分析或希望让 AI 助手直接“阅读”本地文件的用户。虽然生成的内容也具备一定可读性，但其核心优势在于为机器",93400,"2026-04-06T19:52:38",[52,14],{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"ai_summary_zh":66,"readme_en":67,"readme_zh":68,"quickstart_zh":69,"use_case_zh":70,"hero_image_url":71,"owner_login":72,"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":114,"forks":115,"last_commit_at":116,"license":117,"difficulty_score":32,"env_os":118,"env_gpu":119,"env_ram":120,"env_deps":121,"category_tags":125,"github_topics":76,"view_count":32,"oss_zip_url":76,"oss_zip_packed_at":76,"status":17,"created_at":126,"updated_at":127,"faqs":128,"releases":159},8030,"sony\u002Fnnabla","nnabla","Neural Network Libraries","nnabla 是索尼推出的深度学习框架，旨在打通从学术研究、模型开发到生产部署的全流程。它致力于实现“一次编写，处处运行”，无论是桌面电脑、高性能计算集群，还是嵌入式设备与生产服务器，都能流畅执行。\n\n针对开发者在不同硬件环境下适配困难及部署复杂的问题，nnabla 提供了统一的解决方案。其核心采用高效的 C++11 构建，上层封装了灵活易用的 Python API，既支持静态图也支持动态图机制，让用户能根据需求自由选择。独特的“自动前向”功能更允许在运行时动态调整网络结构（如随机丢弃层），极大提升了实验的灵活性。此外，它还配套了 CUDA 加速扩展、神经架构搜索（NAS）、深度强化学习库以及可视化的 Neural Network Console 工具，生态完善。\n\nnnabla 非常适合需要兼顾研究创新与工程落地的算法工程师、科研人员及系统开发者。如果你希望在资源受限的嵌入式设备上部署模型，或追求极致的推理性能，nnabla 是一个值得尝试的专业选择。需注意，目前该项目已进入维护阶段，不再进行活跃的功能开发，但其稳定的核心特性依然适用于现有的生产与研究场景。","***Notice: nnabla is under maintenance phase and we will not be actively developing.***\n\n----\n\n# Neural Network Libraries\n\n[Neural Network Libraries](https:\u002F\u002Farxiv.org\u002Fabs\u002F2102.06725) is a deep learning framework that is intended to be used for research,\ndevelopment and production. We aim to have it running everywhere: desktop PCs, HPC\nclusters, embedded devices and production servers.\n\n\n* [Neural Network Libraries - CUDA extension](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-ext-cuda): An extension library of Neural Network Libraries that allows users to speed-up the computation on CUDA-capable GPUs.\n* [Neural Network Libraries - Examples](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-examples): Working examples of Neural Network Libraries from basic to state-of-the-art.\n* [Neural Network Libraries - C Runtime](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-c-runtime):  Runtime library for inference Neural Network created by Neural Network Libraries.\n* [Neural Network Libraries - NAS](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-nas):  Hardware-aware Neural Architecture Search (NAS) for Neural Network Libraries.\n* [Neural Network Libraries - RL](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-rl):  Deep Reinforcement Learning (RL) library built on top of Neural Network Libraries.\n* [Neural Network Console](https:\u002F\u002Fdl.sony.com\u002F): A Windows GUI app for neural network development.\n\n\n## Installation\n\nInstalling Neural Network Libraries is easy:\n\n```\npip install nnabla\n```\n\nThis installs the CPU version of Neural Network Libraries. GPU-acceleration can be added by installing the CUDA extension with following command.\n```\npip install nnabla-ext-cuda116\n```\nAbove command is for version 11.6 CUDA Toolkit.\n\nThe other supported CUDA packages are listed [here](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fpip_installation_cuda.html#cuda-vs-cudnn-compatibility).\n\nCUDA ver.10.x, ver.9.x, ver.8.x are not supported now.\n\n\nFor more details, see the [installation section](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html) of the documentation.\n\n### Building from Source\n\nSee [Build Manuals](doc\u002Fbuild\u002FREADME.md).\n\n### Running on Docker\nFor details on running on Docker, see the [installation section](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html) of the documentation.\n\n## Features\n\n### Easy, flexible and expressive\n\nThe Python API built on the Neural Network Libraries C++11 core gives you flexibility and\nproductivity. For example, a two layer neural network with classification loss\ncan be defined in the following 5 lines of codes (hyper parameters are enclosed\nby `\u003C>`).\n\n```python\nimport nnabla as nn\nimport nnabla.functions as F\nimport nnabla.parametric_functions as PF\n\nx = nn.Variable(\u003Cinput_shape>)\nt = nn.Variable(\u003Ctarget_shape>)\nh = F.tanh(PF.affine(x, \u003Chidden_size>, name='affine1'))\ny = PF.affine(h, \u003Ctarget_size>, name='affine2')\nloss = F.mean(F.softmax_cross_entropy(y, t))\n```\n\nTraining can be done by:\n\n```python\nimport nnabla.solvers as S\n\n# Create a solver (parameter updater)\nsolver = S.Adam(\u003Csolver_params>)\nsolver.set_parameters(nn.get_parameters())\n\n# Training iteration\nfor n in range(\u003Cnum_training_iterations>):\n    # Setting data from any data source\n    x.d = \u003Cset data>\n    t.d = \u003Cset label>\n    # Initialize gradients\n    solver.zero_grad()\n    # Forward and backward execution\n    loss.forward()\n    loss.backward()\n    # Update parameters by computed gradients\n    solver.update()\n```\n\nThe dynamic computation graph enables flexible runtime network construction.\nNeural Network Libraries can use both paradigms of static and dynamic graphs,\nboth using the same API.\n\n```python\nx.d = \u003Cset data>\nt.d = \u003Cset label>\ndrop_depth = np.random.rand(\u003Cnum_stochastic_layers>) \u003C \u003Clayer_drop_ratio>\nwith nn.auto_forward():\n    h = F.relu(PF.convolution(x, \u003Chidden_size>, (3, 3), pad=(1, 1), name='conv0'))\n    for i in range(\u003Cnum_stochastic_layers>):\n        if drop_depth[i]:\n            continue  # Stochastically drop a layer\n        h2 = F.relu(PF.convolution(x, \u003Chidden_size>, (3, 3), pad=(1, 1), \n                                   name='conv%d' % (i + 1)))\n        h = F.add2(h, h2)\n    y = PF.affine(h, \u003Ctarget_size>, name='classification')\n    loss = F.mean(F.softmax_cross_entropy(y, t))\n# Backward computation (can also be done in dynamically executed graph)\nloss.backward()\n```\n\nYou can differentiate to any order with nn.grad.\n\n```python\nimport nnabla as nn\nimport nnabla.functions as F\nimport numpy as np\n\nx = nn.Variable.from_numpy_array(np.random.randn(2, 2)).apply(need_grad=True)\nx.grad.zero()\ny = F.sin(x)\ndef grad(y, x, n=1):\n    dx = [y]\n    for _ in range(n):\n        dx = nn.grad([dx[0]], [x])\n    return dx[0]\ndnx = grad(y, x, n=10)\ndnx.forward()\nprint(np.allclose(-np.sin(x.d), dnx.d))\ndnx.backward()\nprint(np.allclose(-np.cos(x.d), x.g))\n\n# Show the registry status\nfrom nnabla.backward_functions import show_registry\nshow_registry()\n```\n\n### Command line utility\n\nNeural Network Libraries provides a command line utility `nnabla_cli` for easier use of NNL.\n\n`nnabla_cli` provides following functionality.\n\n- Training, Evaluation or Inference with NNP file.\n- Dataset and Parameter manipulation.\n- File format converter\n  - From ONNX to NNP and NNP to ONNX.\n  - From TensorFlow to NNP and NNP to TensorFlow.\n  - From NNP to TFLite.\n  - From ONNX or NNP to NNB or C source code.\n\nFor more details see [Documentation](doc\u002Fpython\u002Fcommand_line_interface.rst)\n\n\n### Portable and multi-platform\n\n* Python API can be used on Linux and Windows\n* Most of the library code is written in C++14, deployable to embedded devices\n\n### Extensible\n\n* Easy to add new modules like neural network operators and optimizers\n* The library allows developers to add specialized implementations (e.g., for\n  FPGA, ...). For example, we provide CUDA backend as an extension, which gives\n  speed-up by GPU accelerated computation.\n\n### Efficient\n\n* High speed on a single CUDA GPU\n* Memory optimization engine\n* Multiple GPU support\n\n\n## Documentation\n\n\u003Chttps:\u002F\u002Fnnabla.readthedocs.org>\n\n### Getting started\n\n* A number of Jupyter notebook tutorials can be found in the [tutorial](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Ftutorial) folder.\n  We recommend starting from `by_examples.ipynb` for a first\n  working example in Neural Network Libraries and `python_api.ipynb` for an introduction into the\n  Neural Network Libraries API.\n\n* We also provide some more sophisticated examples at [`nnabla-examples`](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-examples) repository.\n\n* C++ API examples are available in [`examples\u002Fcpp`](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp).\n\n\n## Contribution guide\n\nThe technology is rapidly progressing, and researchers and developers often want to add their custom features to a deep learning framework.\nNNabla is really nice in this point. The architecture of Neural Network Libraries is clean and quite simple.\nAlso, you can add new features very easy by the help of our code template generating system.\nSee the following link for details.\n\n* [Contribution guide](CONTRIBUTING.md)\n\n## License & Notice\n\nNeural Network Libraries is provided under the [Apache License Version 2.0](LICENSE) license.\n\nIt also depends on some open source software packages. For more information, see [LICENSES](third_party\u002FLICENSES.md).\n\n## Citation\n\n```\n@misc{hayakawa2021neural,\n      title={Neural Network Libraries: A Deep Learning Framework Designed from Engineers' Perspectives}, \n      author={Takuya Narihira and Javier Alonsogarcia and Fabien Cardinaux and Akio Hayakawa\n              and Masato Ishii and Kazunori Iwaki and Thomas Kemp and Yoshiyuki Kobayashi\n              and Lukas Mauch and Akira Nakamura and Yukio Obuchi and Andrew Shin and Kenji Suzuki\n              and Stephen Tiedmann and Stefan Uhlich and Takuya Yashima and Kazuki Yoshiyama},\n      year={2021},\n      eprint={2102.06725},\n      archivePrefix={arXiv},\n      primaryClass={cs.LG}\n}\n```\n","***注意：nnabla目前处于维护阶段，我们不会进行积极的开发。***\n\n----\n\n# 神经网络库\n\n[神经网络库](https:\u002F\u002Farxiv.org\u002Fabs\u002F2102.06725)是一个深度学习框架，旨在用于研究、开发和生产环境。我们的目标是让其能够在各种平台上运行：台式电脑、高性能计算集群、嵌入式设备以及生产服务器。\n\n\n* [神经网络库 - CUDA扩展](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-ext-cuda)：神经网络库的一个扩展库，允许用户在支持CUDA的GPU上加速计算。\n* [神经网络库 - 示例](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-examples)：从基础到最先进水平的神经网络库示例。\n* [神经网络库 - C运行时](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-c-runtime)：由神经网络库创建的用于推理的运行时库。\n* [神经网络库 - NAS](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-nas)：面向硬件的神经架构搜索（NAS）工具，专为神经网络库设计。\n* [神经网络库 - RL](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-rl)：基于神经网络库构建的深度强化学习（RL）库。\n* [神经网络控制台](https:\u002F\u002Fdl.sony.com\u002F)：一款用于神经网络开发的Windows GUI应用程序。\n\n\n## 安装\n\n安装神经网络库非常简单：\n\n```\npip install nnabla\n```\n\n这将安装神经网络库的CPU版本。若需GPU加速，可通过以下命令安装CUDA扩展：\n```\npip install nnabla-ext-cuda116\n```\n上述命令适用于CUDA Toolkit 11.6版本。\n\n其他支持的CUDA版本列表请参见[这里](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fpip_installation_cuda.html#cuda-vs-cudnn-compatibility)。\n\n目前不支持CUDA 10.x、9.x和8.x版本。\n\n\n更多详细信息，请参阅文档中的[安装部分](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)。\n\n### 从源码编译\n\n请参阅[编译手册](doc\u002Fbuild\u002FREADME.md)。\n\n### 在Docker中运行\n有关在Docker中运行的详细信息，请参阅文档中的[安装部分](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)。\n\n## 特性\n\n### 简单、灵活且表达力强\n\n基于神经网络库C++11核心构建的Python API为您提供了灵活性和高效的工作效率。例如，一个包含分类损失的两层神经网络可以用以下5行代码定义（超参数用`\u003C>`括起）：\n\n```python\nimport nnabla as nn\nimport nnabla.functions as F\nimport nnabla.parametric_functions as PF\n\nx = nn.Variable(\u003Cinput_shape>)\nt = nn.Variable(\u003Ctarget_shape>)\nh = F.tanh(PF.affine(x, \u003Chidden_size>, name='affine1'))\ny = PF.affine(h, \u003Ctarget_size>, name='affine2')\nloss = F.mean(F.softmax_cross_entropy(y, t))\n```\n\n训练过程可以这样进行：\n\n```python\nimport nnabla.solvers as S\n\n# 创建优化器（参数更新器）\nsolver = S.Adam(\u003Csolver_params>)\nsolver.set_parameters(nn.get_parameters())\n\n# 训练迭代\nfor n in range(\u003Cnum_training_iterations>):\n    # 从任何数据源设置数据\n    x.d = \u003Cset data>\n    t.d = \u003Cset label>\n    # 初始化梯度\n    solver.zero_grad()\n    # 前向和反向传播\n    loss.forward()\n    loss.backward()\n    # 根据计算出的梯度更新参数\n    solver.update()\n```\n\n动态计算图使得在网络运行时能够灵活地构建网络结构。神经网络库同时支持静态和动态图两种范式，并且使用相同的API。\n\n```python\nx.d = \u003Cset data>\nt.d = \u003Cset label>\ndrop_depth = np.random.rand(\u003Cnum_stochastic_layers>) \u003C \u003Clayer_drop_ratio>\nwith nn.auto_forward():\n    h = F.relu(PF.convolution(x, \u003Chidden_size>, (3, 3), pad=(1, 1), name='conv0'))\n    for i in range(\u003Cnum_stochastic_layers>):\n        if drop_depth[i]:\n            continue  # 随机跳过某一层\n        h2 = F.relu(PF.convolution(x, \u003Chidden_size>, (3, 3), pad=(1, 1), \n                                   name='conv%d' % (i + 1)))\n        h = F.add2(h, h2)\n    y = PF.affine(h, \u003Ctarget_size>, name='classification')\n    loss = F.mean(F.softmax_cross_entropy(y, t))\n# 反向传播（也可在动态执行的图中完成）\nloss.backward()\n```\n\n您可以使用nn.grad对任意阶导数进行求解。\n\n```python\nimport nnabla as nn\nimport nnabla.functions as F\nimport numpy as np\n\nx = nn.Variable.from_numpy_array(np.random.randn(2, 2)).apply(need_grad=True)\nx.grad.zero()\ny = F.sin(x)\ndef grad(y, x, n=1):\n    dx = [y]\n    for _ in range(n):\n        dx = nn.grad([dx[0]], [x])\n    return dx[0]\ndnx = grad(y, x, n=10)\ndnx.forward()\nprint(np.allclose(-np.sin(x.d), dnx.d))\ndnx.backward()\nprint(np.allclose(-np.cos(x.d), x.g))\n\n# 显示注册表状态\nfrom nnabla.backward_functions import show_registry\nshow_registry()\n```\n\n### 命令行工具\n\n神经网络库提供了一个名为`nnabla_cli`的命令行工具，以方便使用NNL。\n\n`nnabla_cli`提供了以下功能：\n\n- 使用NNP文件进行训练、评估或推理。\n- 数据集和参数操作。\n- 文件格式转换器\n  - 从ONNX转换为NNP，以及从NNP转换为ONNX。\n  - 从TensorFlow转换为NNP，以及从NNP转换为TensorFlow。\n  - 从NNP转换为TFLite。\n  - 从ONNX或NNP转换为NNB或C源代码。\n\n更多详细信息请参阅[文档](doc\u002Fpython\u002Fcommand_line_interface.rst)。\n\n\n### 跨平台且可移植\n\n* Python API可在Linux和Windows上使用。\n* 大部分库代码使用C++14编写，可部署到嵌入式设备上。\n\n### 可扩展性\n\n* 可轻松添加新的模块，如神经网络算子和优化器。\n* 该库允许开发者添加专用实现（例如针对FPGA等）。例如，我们提供了CUDA后端作为扩展，通过GPU加速计算来提升性能。\n\n### 高效\n\n* 在单个CUDA GPU上具有高速性能。\n* 内存优化引擎。\n* 支持多GPU。\n\n\n## 文档\n\n\u003Chttps:\u002F\u002Fnnabla.readthedocs.org>\n\n### 入门指南\n\n* 在[tutorial](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Ftutorial)文件夹中可以找到许多Jupyter Notebook教程。\n  我们建议从`by_examples.ipynb`开始，以获得神经网络库的第一个实际示例；从`python_api.ipynb`开始，则可以了解神经网络库的API。\n\n* 我们还在[`nnabla-examples`](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-examples)仓库中提供了一些更复杂的示例。\n\n* C++ API示例可在[`examples\u002Fcpp`](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp)中找到。\n\n## 贡献指南\n\n技术正在迅速发展，研究人员和开发者常常希望向深度学习框架中添加自定义功能。\nNNabla 在这一点上非常出色。Neural Network Libraries 的架构简洁且相当简单。\n此外，借助我们的代码模板生成系统，您可以非常轻松地添加新功能。\n详情请参阅以下链接。\n\n* [贡献指南](CONTRIBUTING.md)\n\n## 许可与声明\n\nNeural Network Libraries 采用 [Apache License Version 2.0](LICENSE) 许可协议进行授权。\n\n它还依赖于一些开源软件包。更多信息请参阅 [LICENSES](third_party\u002FLICENSES.md)。\n\n## 引用\n\n```\n@misc{hayakawa2021neural,\n      title={Neural Network Libraries: 一款从工程师视角设计的深度学习框架}, \n      author={Takuya Narihira、Javier Alonsogarcia、Fabien Cardinaux、Akio Hayakawa\n             、Masato Ishii、Kazunori Iwaki、Thomas Kemp、Yoshiyuki Kobayashi\n             、Lukas Mauch、Akira Nakamura、Yukio Obuchi、Andrew Shin、Kenji Suzuki\n             、Stephen Tiedmann、Stefan Uhlich、Takuya Yashima、Kazuki Yoshiyama},\n      year={2021},\n      eprint={2102.06725},\n      archivePrefix={arXiv},\n      primaryClass={cs.LG}\n}\n```","# NNabla 快速上手指南\n\n> **注意**：NNabla 目前处于维护阶段，官方不再进行活跃的功能开发，但现有功能仍可稳定用于研究、开发和生产部署。\n\n## 环境准备\n\n*   **操作系统**：Linux 或 Windows。\n*   **Python 版本**：建议 Python 3.6 及以上。\n*   **硬件要求**：\n    *   **CPU 模式**：任意标准 x86 处理器。\n    *   **GPU 模式**：需要 NVIDIA GPU 及对应的 CUDA Toolkit。\n        *   支持 CUDA 11.x (如 `nnabla-ext-cuda116` 对应 CUDA 11.6)。\n        *   **不支持** CUDA 10.x, 9.x, 8.x 等旧版本。\n*   **前置依赖**：确保已安装 `pip` 和基础编译工具（若需从源码构建）。\n\n## 安装步骤\n\n### 1. 安装 CPU 版本\n最基础的安装命令，适用于仅使用 CPU 进行推理或小型模型训练：\n\n```bash\npip install nnabla\n```\n\n*(国内用户建议使用清华源加速安装)*\n```bash\npip install nnabla -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n### 2. 安装 GPU 加速版本\n若需利用 NVIDIA GPU 加速，需额外安装对应的 CUDA 扩展包。以下以 CUDA 11.6 为例：\n\n```bash\npip install nnabla-ext-cuda116\n```\n\n其他支持的 CUDA 版本请参考官方兼容性列表。安装完成后，NNabla 将自动检测并使用 GPU。\n\n## 基本使用\n\nNNabla 提供了简洁的 Python API，支持静态图和动态图混合编程。以下是一个定义简单神经网络并执行一次训练迭代的最小示例。\n\n### 1. 定义网络与损失函数\n\n```python\nimport nnabla as nn\nimport nnabla.functions as F\nimport nnabla.parametric_functions as PF\n\n# 定义输入变量 (占位符)\n# \u003Cinput_shape> 替换为实际形状，例如 (batch_size, channels, height, width)\nx = nn.Variable(\u003Cinput_shape>)\nt = nn.Variable(\u003Ctarget_shape>)\n\n# 构建两层神经网络\nh = F.tanh(PF.affine(x, \u003Chidden_size>, name='affine1'))\ny = PF.affine(h, \u003Ctarget_size>, name='affine2')\n\n# 定义分类损失函数\nloss = F.mean(F.softmax_cross_entropy(y, t))\n```\n\n### 2. 执行训练迭代\n\n```python\nimport nnabla.solvers as S\n\n# 创建优化器 (例如 Adam)\nsolver = S.Adam(\u003Csolver_params>)\nsolver.set_parameters(nn.get_parameters())\n\n# 模拟训练循环\nfor n in range(\u003Cnum_training_iterations>):\n    # 1. 设置数据 (从数据加载器获取)\n    x.d = \u003Cset data>\n    t.d = \u003Cset label>\n    \n    # 2. 清零梯度\n    solver.zero_grad()\n    \n    # 3. 前向传播与反向传播\n    loss.forward()\n    loss.backward()\n    \n    # 4. 更新参数\n    solver.update()\n```\n\n### 3. 动态图特性示例\nNNabla 支持在运行时动态构建计算图（例如随机丢弃层），同时保持高效的执行性能：\n\n```python\nimport numpy as np\n\n# 假设数据已赋值\nx.d = \u003Cset data>\nt.d = \u003Cset label>\n\n# 动态决定网络深度\ndrop_depth = np.random.rand(\u003Cnum_stochastic_layers>) \u003C \u003Clayer_drop_ratio>\n\nwith nn.auto_forward():\n    h = F.relu(PF.convolution(x, \u003Chidden_size>, (3, 3), pad=(1, 1), name='conv0'))\n    for i in range(\u003Cnum_stochastic_layers>):\n        if drop_depth[i]:\n            continue  # 随机跳过某一层\n        h2 = F.relu(PF.convolution(x, \u003Chidden_size>, (3, 3), pad=(1, 1), \n                                   name='conv%d' % (i + 1)))\n        h = F.add2(h, h2)\n    y = PF.affine(h, \u003Ctarget_size>, name='classification')\n    loss = F.mean(F.softmax_cross_entropy(y, t))\n\n# 执行反向传播\nloss.backward()\n```\n\n更多详细教程可参考官方仓库中的 `tutorial` 文件夹或 `nnabla-examples` 项目。","某嵌入式视觉团队正致力于将高精度缺陷检测模型部署到算力受限的工业相机边缘设备上。\n\n### 没有 nnabla 时\n- **跨平台移植困难**：在服务器上用主流框架训练的模型，难以高效迁移至 ARM 架构的嵌入式设备，往往需要重写大量推理代码。\n- **动态结构支持缺失**：尝试实现随机层丢弃（Stochastic Depth）等动态网络结构时，因静态图限制导致代码极其复杂且难以调试。\n- **硬件感知优化繁琐**：缺乏原生支持的神经架构搜索（NAS）工具，手动调整模型以适应特定硬件延迟的过程耗时数周。\n- **训练与推理割裂**：训练环境与最终生产环境的运行时库不一致，导致模型上线后出现精度下降或算子不支持的意外错误。\n\n### 使用 nnabla 后\n- **一次定义，处处运行**：利用 nnabla 统一的 C++ 核心与 Python API，同一套代码可直接在 GPU 集群训练并无缝部署至嵌入式设备的 C 运行时环境。\n- **灵活构建动态图**：借助动态计算图特性，仅用几行代码即可实现随机的层丢弃逻辑，大幅简化了复杂正则化策略的开发。\n- **自动化硬件适配**：通过集成的硬件感知 NAS（nnabla-nas），自动搜索出在目标芯片上速度与精度平衡最优的网络结构，将调优周期缩短至数天。\n- **端到端一致性保障**：从训练到推理全程复用同一套函数库，彻底消除了因环境差异导致的模型表现不一致问题。\n\nnnabla 凭借其“训练即部署”的统一架构，成功打通了从算法研发到边缘落地的最后一公里。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsony_nnabla_cfcd37dc.png","sony","Sony","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fsony_e30df328.png","Sony Group Corporation",null,"https:\u002F\u002Fwww.sony.com\u002Fen\u002F","https:\u002F\u002Fgithub.com\u002Fsony",[80,84,88,92,96,100,104,107,111],{"name":81,"color":82,"percentage":83},"Python","#3572A5",62.2,{"name":85,"color":86,"percentage":87},"C++","#f34b7d",29,{"name":89,"color":90,"percentage":91},"Jupyter Notebook","#DA5B0B",5.8,{"name":93,"color":94,"percentage":95},"Cython","#fedf5b",1.9,{"name":97,"color":98,"percentage":99},"CMake","#DA3434",0.4,{"name":101,"color":102,"percentage":103},"Batchfile","#C1F12E",0.3,{"name":105,"color":106,"percentage":103},"Makefile","#427819",{"name":108,"color":109,"percentage":110},"Dockerfile","#384d54",0.1,{"name":112,"color":113,"percentage":110},"Shell","#89e051",2777,335,"2026-04-02T08:35:37","Apache-2.0","Linux, Windows","非必需（有 CPU 版本）。若需 GPU 加速，需要支持 CUDA 的 NVIDIA 显卡。支持的 CUDA 版本包括 11.6 及其他列出的较新版本；不再支持 CUDA 10.x, 9.x, 8.x。显存大小未说明。","未说明",{"notes":122,"python":120,"dependencies":123},"该工具目前处于维护阶段，不再积极开发。核心库由 C++14 编写，可部署于嵌入式设备。提供命令行工具 nnabla_cli 支持 ONNX、TensorFlow、TFLite 等格式转换。可通过 pip 直接安装 CPU 版本，GPU 版本需单独安装对应的 CUDA 扩展包（如 nnabla-ext-cuda116）。",[124],"nnabla-ext-cuda (可选，用于 GPU 加速)",[14],"2026-03-27T02:49:30.150509","2026-04-16T15:51:16.413507",[129,134,139,144,149,154],{"id":130,"question_zh":131,"answer_zh":132,"source_url":133},35945,"导入 nnabla_ext.cuda 时出现 'ImportError: libcudart.so.11.0: cannot open shared object file' 错误怎么办？","该错误通常由 CUDA 版本不匹配引起。例如，CUDA 11.0 包含 libcusolver.so.10，而 CUDA 11.1 及更高版本包含 libcusolver.so.11。解决方案包括：\n1. 如果系统中有 libcusolver.so.11，创建符号链接指向它。\n2. 降级使用 nnabla v1.26.0 版本。\n3. 在环境中安装匹配的 CUDA 11.0。\n如果是 'libcudnn.so.8' 找不到，请安装 cuDNN 8。详细环境设置可参考官方安装指南：https:\u002F\u002Fnnabla.org\u002Finstall","https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fissues\u002F1063",{"id":135,"question_zh":136,"answer_zh":137,"source_url":138},35946,"运行多 GPU 分布式训练脚本时报错 'Any of [cudnn:float, cuda:float, cpu:float] could not be found' 如何解决？","运行分布式训练需要安装特定的扩展包或使用的 Docker 镜像。请根据 MPI 版本安装以下 pip 包之一：\n- nnabla-ext-cuda110-nccl2-mpi3-1-6 (适用于 mpi 3.1.6 环境)\n- nnabla-ext-cuda110-nccl2-mpi2-1-1 (适用于 mpi 2.1.1 环境)\n或者，直接使用提供的 Docker 镜像运行，例如：\ndocker run --gpus all --rm -it -v $(pwd)\u002Fnnabla-examples:\u002Fopt2 nnabla\u002Fnnabla-ext-cuda-multi-gpu:py36-cuda110-mpi3.1.6-v1.21.0 \u002Fbin\u002Fbash\nmpirun -n 4 python \u002Fopt2\u002Fdistributed\u002Fcifar10-100\u002Fmulti_device_multi_process_classification.py","https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fissues\u002F946",{"id":140,"question_zh":141,"answer_zh":142,"source_url":143},35947,"在 numpy==1.21.4 和 nnabla>=1.29.0 环境下无法导入 nnabla 报错 'numpy.ndarray size changed' 怎么办？","这是由 numpy 版本二进制不兼容引起的。建议升级 nnabla 到最新版本（如 v1.35.1），该版本会自动安装兼容的 numpy (v1.23.5)。操作步骤如下：\n1. 升级 pip: pip install -U pip\n2. 安装最新版 nnabla: pip install nnabla\n安装后可通过 `pip list | grep -e numpy -e nnabla` 确认版本是否为 nnabla 1.35.1 和 numpy 1.23.5。此问题在 v1.33.0 及以上版本已修复。","https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fissues\u002F1107",{"id":145,"question_zh":146,"answer_zh":147,"source_url":148},35948,"NNabla 是否支持 Python 3？如何安装？","是的，NNabla 现已支持 Python 3。用户可以直接通过 pip 在大多数 Linux 和 Windows 64 位机器上安装。相关支持代码已合并到主分支，直接运行 `pip install nnabla` 即可在 Python 3 环境中使用。","https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fissues\u002F2",{"id":150,"question_zh":151,"answer_zh":152,"source_url":153},35949,"哪里可以找到 NNabla 的强化学习（Reinforcement Learning）示例代码？","NNabla 提供了强化学习的示例代码。您可以查看官方的 DQN 示例：https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla-examples\u002Ftree\u002Fmaster\u002Freinforcement_learning\u002Fdqn。此外，社区用户也维护了相关的集合库，例如：https:\u002F\u002Fgithub.com\u002Ftakuseno\u002Fnnabla-drl-collections，其中包含更多模型实现参考。","https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fissues\u002F13",{"id":155,"question_zh":156,"answer_zh":157,"source_url":158},35950,"NNabla 是否支持模型并行（Model Parallelism）？有哪些相关功能？","NNabla 正在逐步完善大模型训练支持。关于模型并行及相关优化技术，可以参考以下发布博客获取详细信息和演示：\n1. 重计算（Recompute）功能介绍：https:\u002F\u002Fblog.nnabla.org\u002Frelease\u002Fv1-20-0\u002F\n2. 内存外计算（Out-of-Core, OoC）功能介绍：https:\u002F\u002Fblog.nnabla.org\u002Frelease\u002Fv1-16-0\u002F\n这些功能旨在帮助在有限显存下训练大型模型。","https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fissues\u002F916",[160,165,170,175,180,185,190,195,200,205,210,215,220,225,230,235,240,245,250,255],{"id":161,"version":162,"summary_zh":163,"released_at":164},281266,"v1.38.0","发布说明-错误修复\n+ [修复输出缺失问题](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1230)\n\n发布说明-构建\n+ [升级 TensorFlow 至 2.14，以避免 NumPy 问题](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1236)\n+ [为 nnabla_cli function_info 添加 -x 参数](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1235)\n+ [将 blockdiag 替换为 graphviz](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1234)\n\n发布说明-文档\n+ [使用 YAML 支持 ReadTheDocs 的 C++ API](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1229)\n\n发布说明-格式转换器\n+ [支持 opset 14 导入器](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1232)\n+ [为 ONNX 导入器添加 Einsum 支持](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1231)\n\n安装最新版 nnabla 的方法如下：\n```bash\npip install nnabla\npip install nnabla-ext-cuda116 # 适用于 CUDA 11.6.2 用户\n```\n\n您也可以从以下选项中安装特定版本的 CUDA 扩展。[更多信息请参阅常见问题解答](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\n+ nnabla-ext-cuda116 (CUDA 11.6.2 x cuDNN 8.4)\n+ nnabla-ext-cuda120 (CUDA 12.0.1 x cuDNN 8.8)\n```bash\npip install nnabla\npip install nnabla_ext_cuda116  # 适用于 CUDA 11.6.2 x cuDNN 8.4 的用户\n```\n\nnnabla 的 CUDA 扩展同时支持集中式和分布式训练。\n以下软件包会检测系统中是否已安装 MPI 和 NCCL2。如果其中任何一个不可用，则分布式训练功能将被禁用。\n\n- nnabla-ext-cuda110  (Ubuntu 20.04 默认)\n- nnabla-ext-cuda116  (Ubuntu 20.04 默认)\n- nnabla-ext-cuda120  (Ubuntu 20.04 默认)\n\n根据您的操作系统或环境，可能还需要进行额外的设置。详情请参阅 Python 软件包安装指南。\n\n[安装指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\n要使用 C++ 推理功能，请按照 C++ 中 MNIST 推理的示例进行操作。\n\n[示例](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\n\n[构建指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\n“nnabla-ext-cuda”软件包目前暂时不可用，不建议使用。请改用 nnabla-ext-cuda110 或 nnabla-ext-cuda116。\n\n以下 nnabla CUDA 扩展包已被弃用，其 PyPI 仓库也已关闭（或即将关闭）：\n\n+ nnabla-ubuntu16\n+ nnabla-ubuntu18\n+ nnabla-ext-cuda\n+ nnabla-ext-cuda80\n+ nnabla-ext-cuda91\n+ nnabla-ext-cuda92\n+ nnabla-ext-cuda100\n+ nnabla-ext-cuda101\n+ nnabla-ext-cuda114\n+ nnabla-ext-cuda90-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu18\n+ nnabla-ext-cuda101-nccl2-ubuntu16\n+ nnabla-ext-cuda101-nccl2-ubuntu18\n+ nnabla-ext-cuda102-nccl2-mpi2-1-1\n+ nnabla-ext-cuda102-nccl2-mpi3-1-6\n+ nnabla-ext-cuda110-nccl2-mpi2-1-1\n+ nnabla-ext-cuda110-nccl2-mpi3-1-6\n\n以下 “nnabla-ext-cuda” Docker 镜像也已被弃用：\n+ py27-cuda92\n+ py36-cuda92\n+ py37-cuda92\n+","2023-12-07T00:57:13",{"id":166,"version":167,"summary_zh":168,"released_at":169},281272,"v1.33.1","发布说明-错误修复\r\n\r\n+ [修复在 CPU 环境下执行 profile 的问题。](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1166)\r\n\r\n请通过以下命令安装最新版本的 nnabla：\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda110 # 适用于 CUDA 11.0.3 用户\r\n```\r\n\r\n您也可以从以下选项中选择特定版本的 CUDA 扩展进行安装。[更多信息请参阅常见问题解答](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\r\n+ nnabla-ext-cuda114 (CUDA 11.4.3 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda110  # 适用于 CUDA 11.0.3 x cuDNN 8.0 用户\r\n```\r\n\r\nnnabla 的 CUDA 扩展同时支持集中式和分布式训练。以下软件包会检测系统中是否已安装 MPI 和 NCCL2。如果其中任何一个不可用，则分布式训练功能将被禁用。\r\n- nnabla-ext-cuda110  (Ubuntu 18.04 默认)\r\n- nnabla-ext-cuda114  (Ubuntu 18.04 默认)\r\n\r\n根据您的操作系统或环境，可能还需要进行额外的设置。详细信息请参阅 Python 包安装指南。\r\n[安装指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\n\r\n若要使用 C++ 推理功能，请按照 MNIST 数据集的 C++ 推理示例进行操作。\r\n[示例](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[构建指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\n\r\n目前，“nnabla-ext-cuda”软件包暂时无法使用，不建议继续使用该包。请改用 nnabla-ext-cuda110 或 nnabla-ext-cuda114。\n\n以下 nnabla CUDA 扩展包已被弃用，并且 PyPI 仓库已经（或将要）关闭。\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda100\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n+ nnabla-ext-cuda102-nccl2-mpi2-1-1\r\n+ nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n+ nnabla-ext-cuda110-nccl2-mpi2-1-1\r\n+ nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\n以下“nnabla-ext-cuda” Docker 镜像也已被弃用。\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n+ py36-cuda100-v1.xx.x\r\n+ py37-cuda100-v1.xx.x\r\n+ py38-cuda100-v1.xx.x\r\n+ py39-cuda100-v1.xx.x\r\n+ py36-cuda102-v1.xx.x\r\n+ py37-cuda102-v1.xx.x\r\n+ py38-cuda102-v1.xx.x\r\n+ py39-cuda102-v1.xx.x\r\n\r\n我们决定将 nnabla 的版本号策略调整为语义化版本控制。此更改自 1.1.0 版本开始实施。\r\n自 1.5.0 版本起，Python 2 将不再受支持。自 1.6.0 版本起，CUDA 8.0 将不再受支持。自 1.14.0 版本起，Python 3.5 将不再受支持。自 1.14.0 版本起，CUDA 9.0 将不再受支持。自 1.26.0 版本起，Python 3.6 将不再受支持。自 1.26.0 版本起，CUDA 10.0 将不再受支持。","2023-02-09T02:51:03",{"id":171,"version":172,"summary_zh":173,"released_at":174},281265,"v1.39.0","发布说明—错误修复\n+ [修复网络扩展器中名称重复的问题](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1239)\n+ [优化器状态恢复的错误修复](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1257)\n\n发布说明—构建\n+ [添加 cctype 头文件以避免在特定平台上出现错误](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1240)\n+ [修复升级 Cython 时的构建错误](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1244)\n+ [限制 Windows 系统上的 Matplotlib 版本](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1259)\n\n发布说明—文档\n+ [更新 nnabla CUDA 12.0 文档](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1243)\n+ [为 nnabla 发布版本更新转换器文档](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1245)\n\n发布说明—格式转换器\n+ [在 ONNX 导入器中，当输入共享数据较大时加快拆分速度](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1249)\n+ [添加输入形状变化时的测试](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1250)\n+ [限制 TensorFlow 版本低于 2.15](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1256)\n\n发布说明—实用工具\n+ [将执行器保存到 graph_def 中](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1241)\n+ [允许设置参数文件格式](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1242)\n+ [在监控器中添加 MLflow 自动日志记录功能](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1251)\n+ [添加代码组织框架](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1252)\n+ [添加 MLflow 模型保存和加载支持](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1253)\n\n安装最新版 nnabla 的方法如下：\n```bash\npip install nnabla\npip install nnabla-ext-cuda116 # 适用于使用 CUDA 11.6.2 的用户\n```\n\n您也可以从以下选项中安装指定版本的 CUDA 扩展。[更多信息请参阅常见问题解答](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\n+ nnabla-ext-cuda116 (CUDA 11.6.2 x cuDNN 8.4)\n+ nnabla-ext-cuda120 (CUDA 12.0.1 x cuDNN 8.8)\n```bash\npip install nnabla\npip install nnabla_ext_cuda116  # 适用于使用 CUDA 11.6.2 x cuDNN 8.4 的用户\n```\n\nnnabla 的 CUDA 扩展同时支持集中式和分布式训练。\n以下软件包会检测系统中是否已安装 MPI 和 NCCL2。如果其中任何一个未安装，则分布式训练功能将被禁用。\n\n- nnabla-ext-cuda110  (Ubuntu 20.04 默认)\n- nnabla-ext-cuda116  (Ubuntu 20.04 默认)\n- nnabla-ext-cuda120  (Ubuntu 20.04 默认)\n\n根据您的操作系统或环境，可能还需要进行额外的配置。有关详细信息，请参阅 Python 软件包安装指南。\n\n[安装指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\n如需使用 C++ 推理功能，请按照 C++ MNIST 推理示例中的演示操作进行。\n\n[演示](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\n\n[构建指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\n“nnabla-ext-cuda”软件包目前暂时不可用，不建议使用。请改用 nnabla-ext-cuda110 或 nnabla-ext-cuda116。\n以下 nnabla CUDA","2024-05-29T05:14:17",{"id":176,"version":177,"summary_zh":178,"released_at":179},281267,"v1.37.0","发布说明-错误修复\n+ [应保存所有执行器属性](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1224)\n+ [修复 ONNX 导入器中的 Slice、MatMul 和 ConstantOfShape](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1223)\n+ [调整 certifi 依赖](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1221)\n+ [为 Windows 全库 wheel 打包 zlib DLL](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1217)\n+ [在构建过程中使 Dockerfile 非交互式](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1216)\n\n发布说明-构建\n+ [将 CentOS 7 替换为 Rocky Linux 8](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1226)\n+ [修复 NumPy 错误](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1225)\n+ [更新 pycodestyle 自动格式化](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1215)\n+ [从 Makefile 中移除 docker inspect](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1214)\n+ [修复 Read the Docs 和 Cython 依赖](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1211)\n+ [升级 Docker 基础镜像至 Ubuntu 20.04](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1208)\n\n发布说明-文档\n+ [更新 nnabla 转换器文档以配合 nnabla 发布](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1210)\n\n发布说明-格式转换器\n+ [更新 TensorFlow 版本并移除 Python 3.7 支持](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1227)\n+ [修复 Unique 的导入器及测试](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1219)\n+ [添加 ONNX 导入器功能：ScatterND、EyeLike、Mod2、BitShift 和 Unique](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1212)\n+ [修复 ONNX 的 range 输入](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1209)\n\n发布说明-实用工具\n+ [支持从 HDF5 文件中读取所有数据类型](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1218)\n\n安装最新版 nnabla 的方法如下：\n```bash\npip install nnabla\npip install nnabla-ext-cuda116 # 适用于 CUDA 11.6.2 用户\n```\n\n您也可以从以下选项中安装特定版本的 CUDA 扩展。[请参阅常见问题解答](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\n+ nnabla-ext-cuda116 (CUDA 11.6.2 x cuDNN 8.4)\n+ nnabla-ext-cuda120 (CUDA 12.0.1 x cuDNN 8.8)\n```bash\npip install nnabla\npip install nnabla_ext_cuda116  # 适用于 CUDA 11.6.2 x cuDNN 8.4 的用户\n```\n\nnnabla CUDA 扩展同时支持集中式和分布式训练。\n以下软件包会检测系统中是否已安装 MPI 和 NCCL2。如果其中任何一个未安装，则分布式训练将被禁用。\n\n- nnabla-ext-cuda110  (Ubuntu 20.04 默认)\n- nnabla-ext-cuda116  (Ubuntu 20.04 默认)\n- nnabla-ext-cuda120  (Ubuntu 20.04 默认)\n\n根据您的操作系统或环境，可能还需要进行额外的设置。有关详细信息，请参阅 Python 包安装指南。\n\n[安装指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\n要使用 C++ 推理功能，请按照 C++ MNIST 推理演示操作。\n\n[演示](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\n\n[构建指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\n“nnabla-ext-cuda”软件包是临时性的。","2023-11-06T05:57:30",{"id":181,"version":182,"summary_zh":183,"released_at":184},281268,"v1.36.0","发布说明-错误修复\n+ [修复张量形状传播错误](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1202)\n+ [修复使用HDF5时的安装失败问题](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1201)\n+ [恢复stft测试，并移除librosa的版本限制](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1195)\n+ [增大插值测试的相对容差](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1186)\n+ [为create_Callback添加dllexport\u002Fimport](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1185)\n\n发布说明-构建\n+ [将nnabla依赖添加到nnabla-converter](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1199)\n+ [为nnabla依赖添加pillow版本限制](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1197)\n+ [更改文档和Colab演示中的默认CUDA版本](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1190)\n+ [通过剥离so库来减小wheel包大小](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1189)\n\n发布说明-文档\n+ [添加.readthedocs.yaml v2](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1203)\n\n发布说明-格式转换器\n+ [可选择ONNX IR版本](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1200)\n+ [升级tf2onnx版本](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1198)\n+ [更改导出器的默认批量大小](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1194)\n\n安装最新版nnabla的方法如下：\n```bash\npip install nnabla\npip install nnabla-ext-cuda116 # 适用于CUDA 11.6.2用户\n```\n\n您也可以从以下选项中安装特定版本的CUDA扩展。[更多信息请参阅常见问题解答](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\n+ nnabla-ext-cuda116 (CUDA 11.6.2 x cuDNN 8.4)\n```bash\npip install nnabla\npip install nnabla_ext_cuda116  # 适用于CUDA 11.6.2 x cuDNN 8.4用户\n```\n\nnnabla CUDA扩展同时支持集中式和分布式训练。\n以下软件包会检测系统中是否已安装MPI和NCCL2。如果其中任何一个不可用，则分布式训练功能将被禁用。\n\n- nnabla-ext-cuda110  (Ubuntu 18.04默认)\n- nnabla-ext-cuda116  (Ubuntu 18.04默认)\n\n根据您的操作系统或环境，可能还需要进行额外的设置。详情请参阅Python包安装指南。\n\n[安装指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\n如需使用C++推理功能，请按照C++ MNIST推理示例进行操作。\n\n[示例](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\n\n[编译指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\n“nnabla-ext-cuda”软件包目前暂时不可用，不建议使用。请改用nnabla-ext-cuda110或nnabla-ext-cuda116。\n\n以下nnabla CUDA扩展包已被弃用，其PyPI仓库也已关闭（或即将关闭）：\n\n+ nnabla-ubuntu16\n+ nnabla-ubuntu18\n+ nnabla-ext-cuda\n+ nnabla-ext-cuda80\n+ nnabla-ext-cuda91\n+ nnabla-ext-cuda92\n+ nnabla-ext-cuda100\n+ nnabla-ext-cuda101\n+ nnabla-ext-cuda114\n+ nnabla-ext-cuda9","2023-07-14T02:40:27",{"id":186,"version":187,"summary_zh":188,"released_at":189},281269,"v1.35.1","发布说明-错误修复\n+ [修复简单数据源打乱问题](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1187)\n\n通过以下命令安装最新版本的 nnabla：\n```bash\npip install nnabla\npip install nnabla-ext-cuda116 # 适用于 CUDA 11.6.2 的用户\n```\n\n您也可以从以下选项中安装特定版本的 CUDA 扩展。[请参阅常见问题解答](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\n+ nnabla-ext-cuda116 (CUDA 11.6.2 x cuDNN 8.4)\n```bash\npip install nnabla\npip install nnabla_ext_cuda116  # 适用于 CUDA 11.6.2 x cuDNN 8.4 的用户\n```\n\nnnabla 的 CUDA 扩展同时支持集中式和分布式训练。\n以下软件包会检测系统上是否已安装 MPI 和 NCCL2。如果其中任何一个不可用，则分布式训练将被禁用。\n\n- nnabla-ext-cuda110  (Ubuntu18.04 默认)\n- nnabla-ext-cuda116  (Ubuntu18.04 默认)\n\n根据您的操作系统或环境，可能还需要进行额外的设置。有关详细信息，请参阅 Python 包安装指南。\n\n[安装指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\n要使用 C++ 推理功能，请按照 MNIST C++ 推理演示中的说明操作。\n\n[演示](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\n\n[构建指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\n“nnabla-ext-cuda”软件包目前暂时不可用，不建议使用。请改用 nnabla-ext-cuda110 或 nnabla-ext-cuda116。\n\n以下 nnabla CUDA 扩展包已被弃用，并且 PyPI 仓库已经（或将）关闭：\n\n+ nnabla-ubuntu16\n+ nnabla-ubuntu18\n+ nnabla-ext-cuda\n+ nnabla-ext-cuda80\n+ nnabla-ext-cuda91\n+ nnabla-ext-cuda92\n+ nnabla-ext-cuda100\n+ nnabla-ext-cuda101\n+ nnabla-ext-cuda114\n+ nnabla-ext-cuda90-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu18\n+ nnabla-ext-cuda101-nccl2-ubuntu16\n+ nnabla-ext-cuda101-nccl2-ubuntu18\n+ nnabla-ext-cuda102-nccl2-mpi2-1-1\n+ nnabla-ext-cuda102-nccl2-mpi3-1-6\n+ nnabla-ext-cuda110-nccl2-mpi2-1-1\n+ nnabla-ext-cuda110-nccl2-mpi3-1-6\n\n以下 “nnabla-ext-cuda” Docker 镜像也已被弃用：\n+ py27-cuda92\n+ py36-cuda92\n+ py37-cuda92\n+ py27-cuda92-v1.0.xx\n+ py36-cuda92-v1.0.xx\n+ py37-cuda92-v1.0.xx\n+ py36-cuda100-v1.xx.x\n+ py37-cuda100-v1.xx.x\n+ py38-cuda100-v1.xx.x\n+ py39-cuda100-v1.xx.x\n+ py36-cuda102-v1.xx.x\n+ py37-cuda102-v1.xx.x\n+ py38-cuda102-v1.xx.x\n+ py39-cuda102-v1.xx.x\n+ py37-cuda114-v1.xx.x\n+ py38-cuda114-v1.xx.x\n+ py39-cuda114-v1.xx.x\n+ py310-cuda114-v1.xx.x\n\n我们决定将 nnabla 的版本控制策略改为语义版本控制。此更改自 1.1.0 版本起生效。\n\nPython 2 将不再受支持，自 v1.5.0 版本起。\nCUDA 8.0 将不再受支持，自 v1.6.0 版本起。\nPython 3.5 将不再受支持，自 v1.14.0 版本起。\nCUDA 9.0 将不再受支持，自…起。","2023-04-25T02:08:35",{"id":191,"version":192,"summary_zh":193,"released_at":194},281270,"v1.35.0","# sony\u002Fnnabla\n\n发布说明-错误修复\n+ [修复 wheel 文件过大问题](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1181)\n\n发布说明-构建\n+ [使用 vs2019 默认路径](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1178)\n+ [在 build-with-docker 中使用用户的 pip 缓存](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1177)\n+ [更改 aarch64 Docker 镜像的基础镜像](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1175)\n+ [当 CUDA≥11.4 时，允许 FFT\u002FIFFT 的 pytest 在 win32 上运行](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1173)\n+ [将 protobuf3.19.4 更新为 protobuf3.19.6](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1172)\n+ [通过修复 librosa 版本解决 stft 测试错误](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1171)\n\n发布说明-文档\n+ [更新 nnabla-doc 构建 requirements.txt](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1180)\n+ [更新 Windows 相关文档](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1179)\n\n发布说明-格式转换器\n+ [修剪常量分支](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1182)\n+ [向 ONNX 导入器添加 8 个算子](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1174)\n\n发布说明-算子层\n+ [Lion 优化器](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1184)\n\n安装最新版 nnabla 的方法如下：\n```bash\npip install nnabla\npip install nnabla-ext-cuda116 # 适用于 CUDA 11.6.2 用户\n```\n\n您也可以从以下选项中安装特定版本的 CUDA 扩展。[请参阅常见问题解答](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\n+ nnabla-ext-cuda116 (CUDA 11.6.2 x cuDNN 8.4)\n```bash\npip install nnabla\npip install nnabla_ext_cuda116  # 适用于 CUDA 11.6.2 x cuDNN 8.4 的用户\n```\n\nnnabla 的 CUDA 扩展同时支持集中式和分布式训练。\n以下软件包会检测系统中是否已安装 MPI 和 NCCL2。如果其中任何一个不可用，则分布式训练将被禁用。\n\n- nnabla-ext-cuda110（Ubuntu 18.04 默认）\n- nnabla-ext-cuda116（Ubuntu 18.04 默认）\n\n根据您的操作系统或环境，可能还需要进行额外的设置。详情请参阅 Python 包安装指南。\n\n[安装指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\n要使用 C++ 推理功能，请按照 C++ MNIST 推理示例中的演示操作。\n\n[演示](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\n\n[构建指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\n“nnabla-ext-cuda” 软件包目前暂时不可用，不建议使用。请改用 nnabla-ext-cuda110 或 nnabla-ext-cuda116。\n\n以下 nnabla CUDA 扩展软件包已被弃用，其 PyPI 仓库也已关闭或即将关闭：\n\n+ nnabla-ubuntu16\n+ nnabla-ubuntu18\n+ nnabla-ext-cuda\n+ nnabla-ext-cuda80\n+ nnabla-ext-cuda91\n+ nnabla-ext-cuda92\n+ nnabla-ext-cuda100\n+ nnabla-ext-cuda101\n+ nnabla-ext-cuda114\n+ nnabla-ext-cuda90-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu18\n+ nnabla","2023-04-06T02:36:03",{"id":196,"version":197,"summary_zh":198,"released_at":199},281271,"v1.34.0","发布说明-错误修复\n+ [检查 SyncBatchNormalization 中内存布局的一致性](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1169)\n+ [抑制子数组释放时的不必要的错误信息](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1161)\n\n发布说明-构建\n+ [修复 cuda11.6 版本中的 alllib 轮子依赖问题](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1167)\n+ [支持 CUDA 11.6](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1164)\n\n发布说明-文档\n+ [更新 BibTeX 和支持的功能](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1168)\n+ [更新 macOS 文档](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1165)\n+ [更新 Linux 文档](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1162)\n\n发布说明-格式转换器\n+ [为 ONNX 导入器添加对 NonZero、NonMaxSuppression、OneHot 和 Resize 的支持](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1159)\n\n发布说明-实用工具\n+ [在每个 epoch 结束时启用切片数据源的打乱](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1160)\n\n通过以下命令安装最新版本的 nnabla：\n```bash\npip install nnabla\npip install nnabla-ext-cuda116 # 适用于 CUDA 11.6.2 用户\n```\n\n您也可以从以下选项中安装特定版本的 CUDA 扩展。[请参阅常见问题解答](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\n+ nnabla-ext-cuda116 (CUDA 11.6.2 x cuDNN 8.4)\n```bash\npip install nnabla\npip install nnabla_ext_cuda116  # 适用于 CUDA 11.6.2 x cuDNN 8.4 的用户\n```\n\nnnabla 的 CUDA 扩展同时支持集中式和分布式训练。以下软件包会检测系统上是否安装了 MPI 和 NCCL2。如果其中任何一个不可用，则分布式训练将被禁用。\n\n- nnabla-ext-cuda110  (Ubuntu 18.04 默认)\n- nnabla-ext-cuda116  (Ubuntu 18.04 默认)\n\n根据您的操作系统或环境，可能还需要进行额外的设置。有关详细信息，请参阅 Python 包安装指南。\n\n[安装指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\n\n要使用 C++ 推理功能，请按照 C++ 中 MNIST 推理的演示操作。\n\n[演示](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\n\n[构建指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\n\n“nnabla-ext-cuda”软件包目前已暂时不可用，不建议使用。请改用 nnabla-ext-cuda110 或 nnabla-ext-cuda116。\n\n以下 nnabla CUDA 扩展软件包已被弃用，并且 PyPI 仓库已经（或将）关闭：\n\n+ nnabla-ubuntu16\n+ nnabla-ubuntu18\n+ nnabla-ext-cuda\n+ nnabla-ext-cuda80\n+ nnabla-ext-cuda91\n+ nnabla-ext-cuda92\n+ nnabla-ext-cuda100\n+ nnabla-ext-cuda101\n+ nnabla-ext-cuda114\n+ nnabla-ext-cuda90-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu18\n+ nnabla-ext-cuda101-nccl2-ubuntu16\n+ nnabla-ext-cuda101-nccl2-ubuntu18\n+ nnabla-ext-cuda102-nccl2-mpi2-1-1\n+ nnabla-ext-cuda102-nccl2-mpi3-1-6\n+ nnabla-ext-cuda110-n","2023-03-01T07:36:21",{"id":201,"version":202,"summary_zh":203,"released_at":204},281273,"v1.33.0","发布说明-错误修复\n+ [修复自动转发模式下内存释放行为中的潜在 bug](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1146)\n+ [动态 numpy 版本](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1139)\n\n发布说明-构建\n+ [将文档和 Makefile 中的默认 Python 版本更新为 3.8](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1155)\n+ [使用 pyenv 构建 Python](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1145)\n+ [添加对 android-ndk-r25b 的支持](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1143)\n\n发布说明-核心\n+ [在 allreduce 回调中添加缩放梯度和保持数据类型选项](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1150)\n\n发布说明-文档\n+ [于 2022 年 12 月 10 日更新 nnabla 转换文档](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1157)\n\n发布说明-格式转换器\n+ [为 proto 图中的 nnp 模型优化添加新子命令](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1154)\n\n发布说明-实用工具\n+ [调整优化器状态的保存和加载功能。](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1156)\n+ [通过优化 HDF5 的使用，支持大于 2GB 的文件](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1153)\n\n安装最新版 nnabla 的方法如下：\n```bash\npip install nnabla\npip install nnabla-ext-cuda110 # 适用于 CUDA 11.0.3 用户\n```\n\n您也可以从以下选项中安装特定版本的 CUDA 扩展。[请参阅常见问题解答](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\n+ nnabla-ext-cuda114 (CUDA 11.4.3 x cuDNN 8.0)\n```bash\npip install nnabla\npip install nnabla_ext_cuda110  # 适用于 CUDA 11.0.3 x cuDNN 8.0 的用户\n```\n\nnnabla CUDA 扩展同时支持集中式和分布式训练。\n以下软件包会检测系统上是否已安装 MPI 和 NCCL2。如果其中任一组件不可用，则分布式训练将被禁用。\n\n- nnabla-ext-cuda110  (Ubuntu 18.04 默认)\n- nnabla-ext-cuda114  (Ubuntu 18.04 默认)\n\n根据您的操作系统或环境，可能还需要进行额外的设置。有关详细信息，请参阅 Python 包安装指南。\n\n[安装指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\n要使用 C++ 推理功能，请按照 C++ MNIST 推理演示操作。\n\n[演示](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\n\n[构建指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\n“nnabla-ext-cuda” 软件包目前暂时不可用，不建议使用。请改用 nnabla-ext-cuda110 或 nnabla-ext-cuda114。\n\n以下 nnabla CUDA 扩展软件包已被弃用，并且 PyPI 仓库已经（或将）关闭：\n\n+ nnabla-ubuntu16\n+ nnabla-ubuntu18\n+ nnabla-ext-cuda\n+ nnabla-ext-cuda80\n+ nnabla-ext-cuda91\n+ nnabla-ext-cuda92\n+ nnabla-ext-cuda100\n+ nnabla-ext-cuda101\n+ nnabla-ext-cuda90-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu18\n+ nnabla-ext-cuda101-nccl2-ubuntu16\n+ nnab","2023-01-23T09:11:53",{"id":206,"version":207,"summary_zh":208,"released_at":209},281274,"v1.32.1","发布说明-构建\n\n+ [修复 wheel 包的命名空间，以通过 twine 检查](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1149)\n\n请通过以下命令安装最新版本的 nnabla：\n```bash\npip install nnabla\npip install nnabla-ext-cuda110 # 适用于 CUDA 11.0.3 的用户\n```\n\n您也可以从以下选项中安装特定版本的 CUDA 扩展。[更多信息请参阅常见问题解答](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\n+ nnabla-ext-cuda110（CUDA 11.0.3 x cuDNN 8.0）\n+ nnabla-ext-cuda114（CUDA 11.4.3 x cuDNN 8.0）\n```bash\npip install nnabla\npip install nnabla_ext_cuda110  # 适用于 CUDA 11.0.3 x cuDNN 8.0 的用户\n```\n\nnnabla 的 CUDA 扩展同时支持集中式和分布式训练。以下软件包会检测系统中是否已安装 MPI 和 NCCL2。如果其中任何一个不可用，则分布式训练将被禁用。\n\n- nnabla-ext-cuda110（Ubuntu 18.04 默认配置）\n- nnabla-ext-cuda114（Ubuntu 18.04 默认配置）\n\n根据您的操作系统或环境，可能还需要进行额外的设置。详细信息请参阅 Python 软件包安装指南。\n\n[安装指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\n\n要使用 C++ 推理功能，请按照 C++ 中 MNIST 推理的示例进行操作。\n\n[示例](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\n\n[构建指南](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\n\n“nnabla-ext-cuda”软件包目前暂时不可用，不建议使用。请改用 nnabla-ext-cuda110 或 nnabla-ext-cuda114。\n\n以下 nnabla CUDA 扩展包已被弃用，并且 PyPI 仓库已经关闭或即将关闭：\n\n+ nnabla-ubuntu16\n+ nnabla-ubuntu18\n+ nnabla-ext-cuda\n+ nnabla-ext-cuda80\n+ nnabla-ext-cuda91\n+ nnabla-ext-cuda92\n+ nnabla-ext-cuda100\n+ nnabla-ext-cuda101\n+ nnabla-ext-cuda90-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu16\n+ nnabla-ext-cuda100-nccl2-ubuntu18\n+ nnabla-ext-cuda101-nccl2-ubuntu16\n+ nnabla-ext-cuda101-nccl2-ubuntu18\n+ nnabla-ext-cuda102-nccl2-mpi2-1-1\n+ nnabla-ext-cuda102-nccl2-mpi3-1-6\n+ nnabla-ext-cuda110-nccl2-mpi2-1-1\n+ nnabla-ext-cuda110-nccl2-mpi3-1-6\n\n以下 “nnabla-ext-cuda” Docker 镜像也已被弃用：\n+ py27-cuda92\n+ py36-cuda92\n+ py37-cuda92\n+ py27-cuda92-v1.0.xx\n+ py36-cuda92-v1.0.xx\n+ py37-cuda92-v1.0.xx\n+ py36-cuda100-v1.xx.x\n+ py37-cuda100-v1.xx.x\n+ py38-cuda100-v1.xx.x\n+ py39-cuda100-v1.xx.x\n+ py36-cuda102-v1.xx.x\n+ py37-cuda102-v1.xx.x\n+ py38-cuda102-v1.xx.x\n+ py39-cuda102-v1.xx.x\n\n我们决定将 nnabla 的版本号策略改为语义化版本控制。此更改自 1.1.0 版本起生效。\n\nPython 2 将不再受支持，自 v1.5.0 版本开始。\nCUDA 8.0 将不再受支持，自 v1.6.0 版本开始。\nPython 3.5 将不再受支持，自 v1.14.0 版本开始。\nCUDA 9.0 将不再受支持，自 v1.14.0 版本开始。\nPython 3.6 将不再受支持，自 v1.26.0 版本开始。\nCUDA 10.0 将不再受支持，自","2023-01-12T10:54:08",{"id":211,"version":212,"summary_zh":213,"released_at":214},281275,"v1.32.0","release-note-bugfix\r\n+ [Avoid net['names'] dict is overwrite](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1135)\r\n+ [Fix save function for the parameters created by narrow](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1130)\r\n+ [Fix the timing of fill and zero evaluations in narrowed array](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1125)\r\n+ [fix crash problem of pipeline](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1124)\r\n+ [fix:skip unnecessary download](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1121)\r\n+ [Not to all-reduce gradients of beta and gamma in sync BN](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1120)\r\n+ [correct index error in generate_cache_dir().](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1114)\r\n\r\nrelease-note-build\r\n+ [Change python version to 3.9.14 in aarch64 environment](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1137)\r\n+ [change pip installation method](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1132)\r\n+ [Support python3.10](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1128)\r\n+ [Added ca-certificates to Dockerfiles](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1119)\r\n+ [fix version mismatch for libarchive.so](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1117)\r\n\r\nrelease-note-core\r\n+ [Release unused memory in auto forward mode](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1127)\r\n\r\nrelease-note-doc\r\n+ [update document for watch dog timeout setting](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1136)\r\n+ [Update nnabla converter document](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1122)\r\n+ [Add callback APIs to doc](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1116)\r\n\r\nrelease-note-examples\r\n+ [Update colab demo scripts and nnabla-ext-cuda version to 114](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1133)\r\n\r\nrelease-note-format-converter\r\n+ [python3.10 converter dependency for onnx and tensorflow](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1134)\r\n+ [Loose converter python packages requirements](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1131)\r\n+ [upgrade onnx to 1.10.0](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1129)\r\n\r\nrelease-note-op-layer\r\n+ [Faster LayerNormCuda::setup_impl (GN and IN as well) by removing unnecessary function creation](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1115)\r\n\r\nrelease-note-utility\r\n+ [Create cache on each distributed node](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1123)\r\n\r\nInstall the latest nnabla by:\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda110 # For CUDA version 11.0.3 users\r\n```\r\n\r\nYou can also install the cuda extension with specific versions from one of the following. [See also FAQ](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\r\n+ nnabla-ext-cuda114 (CUDA 11.4.3 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda110  # For CUDA 11.0.3 x cuDNN 8.0 users\r\n```\r\n\r\nThe nnabla cuda extension supports both centralized and distributed training.\r\nBelow packages detect MPI and NCCL2 installed on your system. If either one\r\nof then is unavailable, distributed training is disabled.\r\n\r\n- nnabla-ext-cuda110  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda114  (Ubuntu18.04 default)\r\n\r\n\r\nAdditional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.\r\n\r\n[Install Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\nTo use C++ inference feature, follow the demonstration on MNIST inference in C++.\r\n\r\n[Demonstration](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[Build Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\nThe \"nnabla-ext-cuda\" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda110, nnabla-ext-cuda114 instead.\r\nThe following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.\r\n\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda100\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n+ nnabla-ext-cuda102-nccl2-mpi2-1-1\r\n+ nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n+ nnabla-ext-cuda110-nccl2-mpi2-1-1\r\n+ nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\n\r\nThe following \"nnabla-ext-cuda\" docker images have been deprecated.\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n+ py36-cuda100-v1.xx.x\r\n+ py37-cuda100-v1.xx.x\r\n+ py38-cuda100-v1.xx.x\r\n+ py39-cuda100-v1.xx.x\r\n+ py36-cuda102-v1.xx.x\r\n+ py37-cuda102-v1.xx.x\r\n+ py38-cuda102-v1.xx.x\r\n+ py39-cuda102-v1.xx.x\r\n\r\nWe've decided to change nnabla's versioning policy to semantic versioning.\r\nThis change has been applied from version 1.1.0.\r\n\r\nPython 2 is no longer be supported from v1.5.0.\r\nCUDA8.0 is no longer be supported from v1.6.0.\r\nPython3.5 is no longer be supported from v1.14.0.\r\nCUDA9.0 is no longer be supported from v1.14.0.\r\nPython3.6 is no longer be","2022-12-07T10:10:49",{"id":216,"version":217,"summary_zh":218,"released_at":219},281276,"v1.31.0","release-note-bugfix\r\n+ [implement drop_last option for slice](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1113)\r\n+ [fix numpy indice error due to numpy version update](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1109)\r\n+ [add code to allow unlink variable](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1105)\r\n\r\nrelease-note-build\r\n+ [add tensorboard version dependency of file format converter](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1110)\r\n+ [build android target with dynamic link](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1099)\r\n\r\nrelease-note-core\r\n+ [Add narrow function for NdArray](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1096)\r\n\r\nrelease-note-format-converter\r\n+ [Add TopK support to ONNX importer](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1111)\r\n+ [Add GlobalMaxPool, RandomNormal, RandomUniform support to ONNX importer](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1103)\r\n+ [Fix issues in ONNX importer](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1102)\r\n+ [restore the CL-initializer in legacy_nnp_graph.py to graph_def.py](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1100)\r\n\r\nrelease-note-op-layer\r\n+ [Error function (erf) implementation (CPU)](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1097)\r\n\r\nInstall the latest nnabla by:\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda110 # For CUDA version 11.0.3 users\r\n```\r\n\r\nYou can also install the cuda extension with specific versions from one of the following. [See also FAQ](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\r\n+ nnabla-ext-cuda114 (CUDA 11.4.3 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda110  # For CUDA 11.0.3 x cuDNN 8.0 users\r\n```\r\n\r\nThe nnabla cuda extension supports both centralized and distributed training.\r\nBelow packages detect MPI and NCCL2 installed on your system. If either one\r\nof then is unavailable, distributed training is disabled.\r\n\r\n- nnabla-ext-cuda110  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda114  (Ubuntu18.04 default)\r\n\r\n\r\nAdditional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.\r\n\r\n[Install Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\nTo use C++ inference feature, follow the demonstration on MNIST inference in C++.\r\n\r\n[Demonstration](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[Build Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\nThe \"nnabla-ext-cuda\" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda110, nnabla-ext-cuda114 instead.\r\nThe following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.\r\n\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda100\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n+ nnabla-ext-cuda102-nccl2-mpi2-1-1\r\n+ nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n+ nnabla-ext-cuda110-nccl2-mpi2-1-1\r\n+ nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\n\r\nThe following \"nnabla-ext-cuda\" docker images have been deprecated.\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n+ py36-cuda100-v1.xx.x\r\n+ py37-cuda100-v1.xx.x\r\n+ py38-cuda100-v1.xx.x\r\n+ py39-cuda100-v1.xx.x\r\n+ py36-cuda102-v1.xx.x\r\n+ py37-cuda102-v1.xx.x\r\n+ py38-cuda102-v1.xx.x\r\n+ py39-cuda102-v1.xx.x\r\n\r\nWe've decided to change nnabla's versioning policy to semantic versioning.\r\nThis change has been applied from version 1.1.0.\r\n\r\nPython 2 is no longer be supported from v1.5.0.\r\nCUDA8.0 is no longer be supported from v1.6.0.\r\nPython3.5 is no longer be supported from v1.14.0.\r\nCUDA9.0 is no longer be supported from v1.14.0.\r\nPython3.6 is no longer be supported from v1.26.0.\r\nCUDA10.0 is no longer be supported from v1.26.0.\r\nCUDA10.2 is no longer be supported from v1.30.0.","2022-10-11T00:19:32",{"id":221,"version":222,"summary_zh":223,"released_at":224},281277,"v1.30.0","release-note-bugfix\r\n+ [fix proto to info.proto in load.py](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1088)\r\n+ [Fix QNN param exists check](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1086)\r\n+ [Fix QNN recorder zero check](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1081)\r\n+ [fix training attribute problem of module](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1078)\r\n\r\nrelease-note-build\r\n+ [Add benchmark for F.transpose()](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1091)\r\n+ [include cuda\u002Fcudnn libraries in wheel](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1090)\r\n+ [allow selection of python to be used when running build_cpplib.bat](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1089)\r\n+ [Update scipy I\u002FF from 1.9.x](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1087)\r\n+ [Skip FFT\u002FIFFT pytests for win32+CUDA114](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1085)\r\n+ [Integrate flatc to converter](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1079)\r\n\r\nrelease-note-format-converter\r\n+ [update version dependency of file format converter](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1095)\r\n+ [pack flatc binary files for windows\u002Flinux\u002Fmac into python wheel](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1084)\r\n+ [add onnx exporter groupnormalization](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1082)\r\n+ [Implementation of onnx import of quantize resize scatter compress shape](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1077)\r\n+ [add shape() function and importer](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1076)\r\n\r\nInstall the latest nnabla by:\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda110 # For CUDA version 11.0.3 users\r\n```\r\n\r\nYou can also install the cuda extension with specific versions from one of the following. [See also FAQ](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda110 (CUDA 11.0.3 x cuDNN 8.0)\r\n+ nnabla-ext-cuda114 (CUDA 11.4.3 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda110  # For CUDA 11.0.3 x cuDNN 8.0 users\r\n```\r\n\r\nThe nnabla cuda extension supports both centralized and distributed training.\r\nBelow packages detect MPI and NCCL2 installed on your system. If either one\r\nof then is unavailable, distributed training is disabled.\r\n\r\n- nnabla-ext-cuda110  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda114  (Ubuntu18.04 default)\r\n\r\n\r\nAdditional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.\r\n\r\n[Install Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\nTo use C++ inference feature, follow the demonstration on MNIST inference in C++.\r\n\r\n[Demonstration](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[Build Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\nThe \"nnabla-ext-cuda\" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda110, nnabla-ext-cuda114 instead.\r\nThe following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.\r\n\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda100\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n+ nnabla-ext-cuda102-nccl2-mpi2-1-1\r\n+ nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n+ nnabla-ext-cuda110-nccl2-mpi2-1-1\r\n+ nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\n\r\nThe following \"nnabla-ext-cuda\" docker images have been deprecated.\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n+ py36-cuda100-v1.xx.x\r\n+ py37-cuda100-v1.xx.x\r\n+ py38-cuda100-v1.xx.x\r\n+ py39-cuda100-v1.xx.x\r\n+ py36-cuda102-v1.xx.x\r\n+ py37-cuda102-v1.xx.x\r\n+ py38-cuda102-v1.xx.x\r\n+ py39-cuda102-v1.xx.x\r\n\r\nWe've decided to change nnabla's versioning policy to semantic versioning.\r\nThis change has been applied from version 1.1.0.\r\n\r\nPython 2 is no longer be supported from v1.5.0.\r\nCUDA8.0 is no longer be supported from v1.6.0.\r\nPython3.5 is no longer be supported from v1.14.0.\r\nCUDA9.0 is no longer be supported from v1.14.0.\r\nPython3.6 is no longer be supported from v1.26.0.\r\nCUDA10.0 is no longer be supported from v1.26.0.\r\nCUDA10.2 is no longer be supported from v1.30.0.\r\n","2022-09-05T09:53:01",{"id":226,"version":227,"summary_zh":228,"released_at":229},281278,"v1.29.0","release-note-bugfix\r\n+ [Derived class from Module wraps all methods for parameter scope](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1069)\r\n+ [fix inf serialization bug](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1067)\r\n+ [save no_image_normalization in executor if exist](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1066)\r\n\r\nrelease-note-build\r\n+ [Adjust protobuf version dependancy on Win\u002FNon-Win platform](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1075)\r\n+ [Update protobuf version](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1074)\r\n+ [fix the problem of tensorboardX support document](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1073)\r\n+ [use numpy 1.20.0 and later](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1072)\r\n+ [Build flatc from latest upstream release](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1068)\r\n\r\nrelease-note-utility\r\n+ [ QNN Converter and QAT Scheduler](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1070)\r\n+ [add warning if parameter file is not supported](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1065)\r\n\r\nInstall the latest nnabla by:\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda102 # For CUDA version 10.2 users\r\n```\r\n\r\nYou can also install the cuda extension with specific versions from one of the following. [See also FAQ](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda102 (CUDA 10.2 x cuDNN 8.0)\r\n+ nnabla-ext-cuda110 (CUDA 11.0 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda102  # For CUDA 10.2 x cuDNN 8.0 users\r\n```\r\n\r\nFor distributed training, You need to install the correct package that\r\nmatches the version of MPI installed on your system.\r\n\r\nWe prepared following packages.\r\n\r\n- nnabla-ext-cuda102-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n- nnabla-ext-cuda110-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\nIf you want to use a version of MPI not listed above, you need to build it from the source.\r\n\r\n\r\nAdditional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.\r\n\r\n[Install Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\nTo use C++ inference feature, follow the demonstration on MNIST inference in C++.\r\n\r\n[Demonstration](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[Build Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\nThe \"nnabla-ext-cuda\" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda110, nnabla-ext-cuda102 instead.\r\nThe following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.\r\n\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda100\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n\r\nThe following \"nnabla-ext-cuda\" docker images have been deprecated.\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n+ py36-cuda100\r\n\r\nWe've decided to change nnabla's versioning policy to semantic versioning.\r\nThis change has been applied from version 1.1.0.\r\n\r\nPython 2 is no longer be supported from v1.5.0.\r\nCUDA8.0 is no longer be supported from v1.6.0.\r\nPython3.5 is no longer be supported from v1.14.0.\r\nCUDA9.0 is no longer be supported from v1.14.0.\r\nPython3.6 is no longer be supported from v1.26.0.\r\nCUDA10.0 is no longer be supported from v1.26.0.\r\n","2022-06-20T00:38:12",{"id":231,"version":232,"summary_zh":233,"released_at":234},281279,"v1.28.0","release-note-bugfix\r\n+ [fix yolov2 load error](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1051)\r\n+ [Fix scatter_add for recompute](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1050)\r\n+ [fix segmetation load bug](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1049)\r\n+ [Fix param dtype translate in python test.](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1047)\r\n+ [make varaible's slice consistent between function call and build-in o…](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1045)\r\n\r\nrelease-note-build\r\n+ [Fix flatc build issue](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1060)\r\n\r\nrelease-note-cpp-api\r\n+ [support save parameters to buffer](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1062)\r\n+ [Add the support of loading parameter from buffer](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1053)\r\n\r\nrelease-note-doc\r\n+ [Update nnabla convert doc at 20220411](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1057)\r\n\r\nrelease-note-format-converter\r\n+ [implement version-down converter of nnp to nnp](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1056)\r\n+ [feat: extend onnx importer opset12](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1046)\r\n\r\nrelease-note-utility\r\n+ [add a new GraphConverter to support weight pruning](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1058)\r\n\r\nInstall the latest nnabla by:\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda102 # For CUDA version 10.2 users\r\n```\r\n\r\nYou can also install the cuda extension with specific versions from one of the following. [See also FAQ](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda102 (CUDA 10.2 x cuDNN 8.0)\r\n+ nnabla-ext-cuda110 (CUDA 11.0 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda102  # For CUDA 10.2 x cuDNN 8.0 users\r\n```\r\n\r\nFor distributed training, You need to install the correct package that\r\nmatches the version of MPI installed on your system.\r\n\r\nWe prepared following packages.\r\n\r\n- nnabla-ext-cuda102-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n- nnabla-ext-cuda110-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\nIf you want to use a version of MPI not listed above, you need to build it from the source.\r\n\r\n\r\nAdditional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.\r\n\r\n[Install Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\nTo use C++ inference feature, follow the demonstration on MNIST inference in C++.\r\n\r\n[Demonstration](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[Build Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\nThe \"nnabla-ext-cuda\" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda110, nnabla-ext-cuda102 instead.\r\nThe following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.\r\n\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda100\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n\r\nThe following \"nnabla-ext-cuda\" docker images have been deprecated.\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n+ py36-cuda100\r\n\r\nWe've decided to change nnabla's versioning policy to semantic versioning.\r\nThis change has been applied from version 1.1.0.\r\n\r\nPython 2 is no longer be supported from v1.5.0.\r\nCUDA8.0 is no longer be supported from v1.6.0.\r\nPython3.5 is no longer be supported from v1.14.0.\r\nCUDA9.0 is no longer be supported from v1.14.0.\r\nPython3.6 is no longer be supported from v1.26.0.\r\nCUDA10.0 is no longer be supported from v1.26.0.\r\n","2022-05-02T00:27:26",{"id":236,"version":237,"summary_zh":238,"released_at":239},281280,"v1.27.0","release-note-bugfix\r\n+ [Fix object detector models](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F989)\r\n\r\nrelease-note-build\r\n+ [arm64 support](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1039)\r\n\r\nrelease-note-core\r\n+ [Fused weight decay into Solver update \u002F Fix SgdW \u002F Add Lamb solver](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1038)\r\n+ [Enable customizing std types and functions with memory allocation](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1037)\r\n\r\nrelease-note-format-converter\r\n+ [Improve bnfolding converter](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1042)\r\n\r\nrelease-note-op-layer\r\n+ [Add batch cholesky function](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1041)\r\n\r\nInstall the latest nnabla by:\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda102 # For CUDA version 10.2 users\r\n```\r\n\r\nYou can also install the cuda extension with specific versions from one of the following. [See also FAQ](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda102 (CUDA 10.2 x cuDNN 8.0)\r\n+ nnabla-ext-cuda110 (CUDA 11.0 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda102  # For CUDA 10.2 x cuDNN 8.0 users\r\n```\r\n\r\nFor distributed training, You need to install the correct package that\r\nmatches the version of MPI installed on your system.\r\n\r\nWe prepared following packages.\r\n\r\n- nnabla-ext-cuda102-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n- nnabla-ext-cuda110-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\nIf you want to use a version of MPI not listed above, you need to build it from the source.\r\n\r\n\r\nAdditional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.\r\n\r\n[Install Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\nTo use C++ inference feature, follow the demonstration on MNIST inference in C++.\r\n\r\n[Demonstration](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[Build Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\nThe \"nnabla-ext-cuda\" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda110, nnabla-ext-cuda102 instead.\r\nThe following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.\r\n\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda100\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n\r\nThe following \"nnabla-ext-cuda\" docker images have been deprecated.\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n+ py36-cuda100\r\n\r\nWe've decided to change nnabla's versioning policy to semantic versioning.\r\nThis change has been applied from version 1.1.0.\r\n\r\nPython 2 is no longer be supported from v1.5.0.\r\nCUDA8.0 is no longer be supported from v1.6.0.\r\nPython3.5 is no longer be supported from v1.14.0.\r\nCUDA9.0 is no longer be supported from v1.14.0.\r\nPython3.6 is no longer be supported from v1.26.0.\r\nCUDA10.0 is no longer be supported from v1.26.0.\r\n","2022-04-11T00:57:34",{"id":241,"version":242,"summary_zh":243,"released_at":244},281281,"v1.26.0","release-note-build\r\n+ [upgrade tensorflow from 2.5.1 to 2.7.x](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1036)\r\n+ [build with VS2019](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1034)\r\n+ [drop support python3.6 and CUDA-cuDNN 10.0\u002F7](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1031)\r\n+ [Feature\u002F20211203 unify dockerfiles](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1028)\r\n\r\nrelease-note-core\r\n+ [restrict condition with dst for communicator reduce](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1032)\r\n+ [Add computation graph active inputs handling.](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1027)\r\n\r\nrelease-note-op-layer\r\n+ [Add Pad to FusedConvolution ](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1035)\r\n\r\nrelease-note-utility\r\n+ [force epoch begin and end callback be called from main thread](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1033)\r\n+ [encode unusual characters in path name](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1030)\r\n\r\nInstall the latest nnabla by:\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda102 # For CUDA version 10.2 users\r\n```\r\n\r\nYou can also install the cuda extension with specific versions from one of the following. [See also FAQ](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda102 (CUDA 10.2 x cuDNN 8.0)\r\n+ nnabla-ext-cuda110 (CUDA 11.0 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda102  # For CUDA 10.2 x cuDNN 8.0 users\r\n```\r\n\r\nFor distributed training, You need to install the correct package that\r\nmatches the version of MPI installed on your system.\r\n\r\nWe prepared following packages.\r\n\r\n- nnabla-ext-cuda102-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n- nnabla-ext-cuda110-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\nIf you want to use a version of MPI not listed above, you need to build it from the source.\r\n\r\n\r\nAdditional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.\r\n\r\n[Install Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\nTo use C++ inference feature, follow the demonstration on MNIST inference in C++.\r\n\r\n[Demonstration](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[Build Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\nThe \"nnabla-ext-cuda\" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda110, nnabla-ext-cuda102 instead.\r\nThe following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.\r\n\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda100\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n\r\nThe following \"nnabla-ext-cuda\" docker images have been deprecated.\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n+ py36-cuda100\r\n\r\nWe've decided to change nnabla's versioning policy to semantic versioning.\r\nThis change has been applied from version 1.1.0.\r\n\r\nPython 2 is no longer be supported from v1.5.0.\r\nCUDA8.0 is no longer be supported from v1.6.0.\r\nPython3.5 is no longer be supported from v1.14.0.\r\nCUDA9.0 is no longer be supported from v1.14.0.\r\nPython3.6 is no longer be supported from v1.26.0.\r\nCUDA10.0 is no longer be supported from v1.26.0.\r\n","2022-03-18T06:12:38",{"id":246,"version":247,"summary_zh":248,"released_at":249},281282,"v1.25.0","release-note-bugfix\r\n+ [Fix FusedBN not to fall back to cpu implementation always](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1023)\r\n+ [Fix illegal access in deformable convolution](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1018)\r\n\r\nrelease-note-build\r\n+ [Make Dockerfile.document depends on doc\u002Frequirements.txt](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1015)\r\n\r\nrelease-note-examples\r\n+ [make corresponding modification for the files in reconstruction directories](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1020)\r\n\r\nrelease-note-op-layer\r\n+ [negative axis support for all functions](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1021)\r\n\r\nInstall the latest nnabla by:\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda102 # For CUDA version 10.2 users\r\n```\r\n\r\nYou can also install the cuda extension with specific versions from one of the following. [See also FAQ](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.6)\r\n+ nnabla-ext-cuda102 (CUDA 10.2 x cuDNN 8.0)\r\n+ nnabla-ext-cuda110 (CUDA 11.0 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda102  # For CUDA 10.2 x cuDNN 8.0 users\r\n```\r\n\r\nFor distributed training, You need to install the correct package that\r\nmatches the version of MPI installed on your system.\r\n\r\nWe prepared following packages.\r\n\r\n- nnabla-ext-cuda100-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda100-nccl2-mpi3-1-6\r\n- nnabla-ext-cuda102-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n- nnabla-ext-cuda110-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\nIf you want to use a version of MPI not listed above, you need to build it from the source.\r\n\r\n\r\nAdditional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.\r\n\r\n[Install Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\nTo use C++ inference feature, follow the demonstration on MNIST inference in C++.\r\n\r\n[Demonstration](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[Build Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\nThe \"nnabla-ext-cuda\" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda110, nnabla-ext-cuda102, nnabla-ext-cuda100 instead.\r\nThe following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.\r\n\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n\r\nThe following \"nnabla-ext-cuda\" docker images have been deprecated.\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n\r\nWe've decided to change nnabla's versioning policy to semantic versioning.\r\nThis change has been applied from version 1.1.0.\r\n\r\nPython 2 is no longer be supported from v1.5.0.\r\nCUDA8.0 is no longer be supported from v1.6.0.\r\nPython3.5 is no longer be supported from v1.14.0.\r\nCUDA9.0 is no longer be supported from v1.14.0.\r\n","2022-01-26T00:19:08",{"id":251,"version":252,"summary_zh":253,"released_at":254},281283,"v1.24.0","release-note-bugfix\r\n+ [do not remove identity layer if it is not created by expend control](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1009)\r\n+ [Remove the argument \"output_mask\" from Dropout](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1006)\r\n+ [fix multithread potential issue](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1005)\r\n+ [Fix recomputation for the function which requires output data for backward computation](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F998)\r\n\r\nrelease-note-build\r\n+ [Add PYTEST_OPTS for pytest parallel execution](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F997)\r\n+ [support python3.9](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F992)\r\n\r\nrelease-note-doc\r\n+ [Update QuantizeLinear&QuantizeLinear support status](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1012)\r\n+ [improve document for watch dog](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1008)\r\n+ [merge cpp-api document](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1007)\r\n+ [revise readthedocs document](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1002)\r\n+ [Update nnabla convert doc at 20211126](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F996)\r\n\r\nrelease-note-format-converter\r\n+ [limit tensorflow version to 2.5.1](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1001)\r\n\r\nrelease-note-op-layer\r\n+ [Support broadcast in instance norm kernel to improve performance ](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F1000)\r\n+ [Make ISTFT consistent with PyTorch implementation (NOLA condition)](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F999)\r\n+ [Add linspace function](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F995)\r\n+ [Optimize CumProd\u002FCumSum](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F993)\r\n\r\nrelease-note-utility\r\n+ [Improve recomputation API](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F994)\r\n\r\nInstall the latest nnabla by:\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda102 # For CUDA version 10.2 users\r\n```\r\n\r\nYou can also install the cuda extension with specific versions from one of the following. [See also FAQ](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.6)\r\n+ nnabla-ext-cuda102 (CUDA 10.2 x cuDNN 8.0)\r\n+ nnabla-ext-cuda110 (CUDA 11.0 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda102  # For CUDA 10.2 x cuDNN 8.0 users\r\n```\r\n\r\nFor distributed training, You need to install the correct package that\r\nmatches the version of MPI installed on your system.\r\n\r\nWe prepared following packages.\r\n\r\n- nnabla-ext-cuda100-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda100-nccl2-mpi3-1-6\r\n- nnabla-ext-cuda102-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n- nnabla-ext-cuda110-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\nIf you want to use a version of MPI not listed above, you need to build it from the source.\r\n\r\n\r\nAdditional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.\r\n\r\n[Install Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\nTo use C++ inference feature, follow the demonstration on MNIST inference in C++.\r\n\r\n[Demonstration](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[Build Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\nThe \"nnabla-ext-cuda\" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda110, nnabla-ext-cuda102, nnabla-ext-cuda100 instead.\r\nThe following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.\r\n\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n\r\nThe following \"nnabla-ext-cuda\" docker images have been deprecated.\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n\r\nWe've decided to change nnabla's versioning policy to semantic versioning.\r\nThis change has been applied from version 1.1.0.\r\n\r\nPython 2 is no longer be supported from v1.5.0.\r\nCUDA8.0 is no longer be supported from v1.6.0.\r\nPython3.5 is no longer be supported from v1.14.0.\r\nCUDA9.0 is no longer be supported from v1.14.0.\r\n","2021-12-27T03:07:05",{"id":256,"version":257,"summary_zh":258,"released_at":259},281284,"v1.23.0","release-note-bugfix\r\n+ [feat: update image_utils for updated opencv-python package.](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F985)\r\n+ [fix remove_and_rewire causes set_variable disconnected issue](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F981)\r\n\r\nrelease-note-build\r\n+ [auto-format-update](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F988)\r\n+ [Skip test_warp_by_grid.py](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F987)\r\n+ [Feature\u002F20211026 pep600 centos7](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F986)\r\n+ [Feature\u002F20211014 pytest parallel exec](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F983)\r\n+ [Apply new pyyaml loader()](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F980)\r\n\r\nrelease-note-core\r\n+ [Add Module.zero_grad](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F990)\r\n\r\nrelease-note-format-converter\r\n+ [feat: add onnx opset13 support.](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Fpull\u002F984)\r\n\r\nInstall the latest nnabla by:\r\n```\r\npip install nnabla\r\npip install nnabla-ext-cuda102 # For CUDA version 10.2 users\r\n```\r\n\r\nYou can also install the cuda extension with specific versions from one of the following. [See also FAQ](https:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstall_on_linux.html#faq)\r\n+ nnabla-ext-cuda100 (CUDA 10.0 x cuDNN 7.6)\r\n+ nnabla-ext-cuda102 (CUDA 10.2 x cuDNN 8.0)\r\n+ nnabla-ext-cuda110 (CUDA 11.0 x cuDNN 8.0)\r\n```\r\npip install nnabla\r\npip install nnabla_ext_cuda102  # For CUDA 10.2 x cuDNN 8.0 users\r\n```\r\n\r\nFor distributed training, You need to install the correct package that\r\nmatches the version of MPI installed on your system.\r\n\r\nWe prepared following packages.\r\n\r\n- nnabla-ext-cuda100-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda100-nccl2-mpi3-1-6\r\n- nnabla-ext-cuda102-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda102-nccl2-mpi3-1-6\r\n- nnabla-ext-cuda110-nccl2-mpi2-1-1  (Ubuntu18.04 default)\r\n- nnabla-ext-cuda110-nccl2-mpi3-1-6\r\n\r\nIf you want to use a version of MPI not listed above, you need to build it from the source.\r\n\r\n\r\nAdditional setup may be required depending on your OS or environment. Please check Python Package Installation Guide for details.\r\n\r\n[Install Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Finstallation.html)\r\nTo use C++ inference feature, follow the demonstration on MNIST inference in C++.\r\n\r\n[Demonstration](https:\u002F\u002Fgithub.com\u002Fsony\u002Fnnabla\u002Ftree\u002Fmaster\u002Fexamples\u002Fcpp\u002Fmnist_runtime)\r\n\r\n[Build Guide](http:\u002F\u002Fnnabla.readthedocs.io\u002Fen\u002Flatest\u002Fpython\u002Fbuild_on_linux_with_dt.html)\r\nThe \"nnabla-ext-cuda\" package is temporarily unavailable. Use of this package is not recommended. Please install nnabla-ext-cuda110, nnabla-ext-cuda102, nnabla-ext-cuda100 instead.\r\nThe following nnabla CUDA extension packages have been deprecated and the PyPi repository has been (or going to be) closed.\r\n\r\n+ nnabla-ubuntu16\r\n+ nnabla-ubuntu18\r\n+ nnabla-ext-cuda\r\n+ nnabla-ext-cuda80\r\n+ nnabla-ext-cuda91\r\n+ nnabla-ext-cuda92\r\n+ nnabla-ext-cuda101\r\n+ nnabla-ext-cuda90-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu16\r\n+ nnabla-ext-cuda100-nccl2-ubuntu18\r\n+ nnabla-ext-cuda101-nccl2-ubuntu16\r\n+ nnabla-ext-cuda101-nccl2-ubuntu18\r\n\r\nThe following \"nnabla-ext-cuda\" docker images have been deprecated.\r\n+ py27-cuda92\r\n+ py36-cuda92\r\n+ py37-cuda92\r\n+ py27-cuda92-v1.0.xx\r\n+ py36-cuda92-v1.0.xx\r\n+ py37-cuda92-v1.0.xx\r\n\r\nWe've decided to change nnabla's versioning policy to semantic versioning.\r\nThis change has been applied from version 1.1.0.\r\n\r\nPython 2 is no longer be supported from v1.5.0.\r\nCUDA8.0 is no longer be supported from v1.6.0.\r\nPython3.5 is no longer be supported from v1.14.0.\r\nCUDA9.0 is no longer be supported from v1.14.0.\r\n","2021-11-25T10:43:29"]