[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-keras-team--keras-hub":3,"tool-keras-team--keras-hub":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":79,"owner_email":80,"owner_twitter":79,"owner_website":81,"owner_url":82,"languages":83,"stars":95,"forks":96,"last_commit_at":97,"license":98,"difficulty_score":23,"env_os":99,"env_gpu":99,"env_ram":99,"env_deps":100,"category_tags":108,"github_topics":109,"view_count":10,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":120,"updated_at":121,"faqs":122,"releases":151},563,"keras-team\u002Fkeras-hub","keras-hub","Pretrained model hub for Keras 3.","KerasHub 是专为 Keras 3 打造的预训练模型中心，汇集了主流模型架构与高质量权重，支持文本、图像及音频任务。它解决了从零训练耗时耗力、复现困难的问题，让开发者能快速获取基座模型进行微调。\n\n作为 Keras API 的扩展，熟悉 Keras 的用户可无缝上手。它特别适合深度学习开发者、算法工程师及研究人员。最大亮点在于支持多后端协同，同一模型定义可在 JAX、TensorFlow 和 PyTorch 间切换，并原生支持 GPU 或 TPU 微调。配合内置 PEFT 技术，无论是单卡实验还是分布式训练都游刃有余。安装简便，仅需 pip 命令即可引入，是构建高效 AI 应用的得力助手。","# KerasHub: Multi-framework Pretrained Models\n[![](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fworkflows\u002FTests\u002Fbadge.svg?branch=master)](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Factions?query=workflow%3ATests+branch%3Amaster)\n![Python](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-v3.11.0+-success.svg)\n[![Kaggle Models](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FKaggle-Models-brightgreen?colorA=0099ff)](https:\u002F\u002Fwww.kaggle.com\u002Forganizations\u002Fkeras\u002Fmodels)\n[![contributions welcome](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fcontributions-welcome-brightgreen.svg?style=flat)](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues)\n\n> [!IMPORTANT]\n> 📢 KerasNLP is now KerasHub! 📢 Read\n> [the announcement](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues\u002F1831).\n\n**KerasHub** is a pretrained modeling library that aims to be simple, flexible,\nand fast. The library provides [Keras 3](https:\u002F\u002Fkeras.io\u002Fkeras_3\u002F)\nimplementations of popular model architectures, paired with a collection of\npretrained checkpoints available on [Kaggle Models](https:\u002F\u002Fwww.kaggle.com\u002Forganizations\u002Fkeras\u002Fmodels).\nModels can be used with text, image, and audio data for generation, classification,\nand many other built in tasks.\n\nKerasHub is an extension of the core Keras API; KerasHub components are provided\nas `Layer` and `Model` implementations. If you are  familiar with Keras,\ncongratulations! You already understand most of KerasHub.\n\nAll models support JAX, TensorFlow, and PyTorch from a single model\ndefinition and can be fine-tuned on GPUs and TPUs out of the box. Models can\nbe trained on individual accelerators with built-in PEFT techniques, or\nfine-tuned at scale with model and data parallel training. See our\n[Getting Started guide](https:\u002F\u002Fkeras.io\u002Fguides\u002Fkeras_hub\u002Fgetting_started)\nto start learning our API.\n\n## Quick Links\n\n### For everyone\n\n- [Home page](https:\u002F\u002Fkeras.io\u002Fkeras_hub)\n- [Getting started](https:\u002F\u002Fkeras.io\u002Fkeras_hub\u002Fgetting_started)\n- [Guides](https:\u002F\u002Fkeras.io\u002Fkeras_hub\u002Fguides)\n- [API documentation](https:\u002F\u002Fkeras.io\u002Fkeras_hub\u002Fapi)\n- [Pre-trained models](https:\u002F\u002Fkeras.io\u002Fkeras_hub\u002Fpresets\u002F)\n\n### For contributors\n\n- [Call for Contributions](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues\u002F1835)\n- [Roadmap](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues\u002F1836)\n- [Contributing Guide](CONTRIBUTING.md)\n- [Style Guide](STYLE_GUIDE.md)\n- [API Design Guide](API_DESIGN_GUIDE.md)\n\n## Quickstart\n\nChoose a backend:\n\n```python\nimport os\nos.environ[\"KERAS_BACKEND\"] = \"jax\"  # Or \"tensorflow\" or \"torch\"!\n```\n\nImport KerasHub and other libraries:\n\n```python\nimport keras\nimport keras_hub\nimport numpy as np\nimport tensorflow_datasets as tfds\n```\n\nLoad a resnet model and use it to predict a label for an image:\n\n```python\nclassifier = keras_hub.models.ImageClassifier.from_preset(\n    \"resnet_50_imagenet\",\n    activation=\"softmax\",\n)\nurl = \"https:\u002F\u002Fupload.wikimedia.org\u002Fwikipedia\u002Fcommons\u002Fa\u002Faa\u002FCalifornia_quail.jpg\"\npath = keras.utils.get_file(origin=url)\nimage = keras.utils.load_img(path)\npreds = classifier.predict(np.array([image]))\nprint(keras_hub.utils.decode_imagenet_predictions(preds))\n```\n\nLoad a Bert model and fine-tune it on IMDb movie reviews:\n\n```python\nclassifier = keras_hub.models.TextClassifier.from_preset(\n    \"bert_base_en_uncased\",\n    activation=\"softmax\",\n    num_classes=2,\n)\nimdb_train, imdb_test = tfds.load(\n    \"imdb_reviews\",\n    split=[\"train\", \"test\"],\n    as_supervised=True,\n    batch_size=16,\n)\nclassifier.fit(imdb_train, validation_data=imdb_test)\npreds = classifier.predict([\"What an amazing movie!\", \"A total waste of time.\"])\nprint(preds)\n```\n\n## Installation\n\nTo install the latest KerasHub release with Keras 3, simply run:\n\n```\npip install --upgrade keras-hub\n```\n\nTo install the latest nightly changes for both KerasHub and Keras, you can use\nour nightly package.\n\n```\npip install --upgrade keras-hub-nightly\n```\n\nCurrently, installing KerasHub will always pull in TensorFlow for use of the\n`tf.data` API for preprocessing. When pre-processing with `tf.data`, training\ncan still happen on any backend.\n\nVisit the [core Keras getting started page](https:\u002F\u002Fkeras.io\u002Fgetting_started\u002F)\nfor more information on installing Keras 3, accelerator support, and\ncompatibility with different frameworks.\n\n## Configuring your backend\n\nIf you have Keras 3 installed in your environment (see installation above),\nyou can use KerasHub with any of JAX, TensorFlow and PyTorch. To do so, set the\n`KERAS_BACKEND` environment variable. For example:\n\n```shell\nexport KERAS_BACKEND=jax\n```\n\nOr in Colab, with:\n\n```python\nimport os\nos.environ[\"KERAS_BACKEND\"] = \"jax\"\n\nimport keras_hub\n```\n\n> [!IMPORTANT]\n> Make sure to set the `KERAS_BACKEND` **before** importing any Keras libraries;\n> it will be used to set up Keras when it is first imported.\n\n## Compatibility\n\nWe follow [Semantic Versioning](https:\u002F\u002Fsemver.org\u002F), and plan to\nprovide backwards compatibility guarantees both for code and saved models built\nwith our components. While we continue with pre-release `0.y.z` development, we\nmay break compatibility at any time and APIs should not be considered stable.\n\n## Disclaimer\n\nKerasHub provides access to pre-trained models via the `keras_hub.models` API.\nThese pre-trained models are provided on an \"as is\" basis, without warranties\nor conditions of any kind. The following underlying models are provided by third\nparties, and subject to separate licenses:\nBART, BLOOM, DeBERTa, DistilBERT, GPT-2, Llama, Mistral, OPT, RoBERTa, Whisper,\nand XLM-RoBERTa.\n\n## Citing KerasHub\n\nIf KerasHub helps your research, we appreciate your citations.\nHere is the BibTeX entry:\n\n```bibtex\n@misc{kerashub2024,\n  title={KerasHub},\n  author={Watson, Matthew, and  Chollet, Fran\\c{c}ois and Sreepathihalli,\n  Divyashree, and Saadat, Samaneh and Sampath, Ramesh, and Rasskin, Gabriel and\n  and Zhu, Scott and Singh, Varun and Wood, Luke and Tan, Zhenyu and Stenbit,\n  Ian and Qian, Chen, and Bischof, Jonathan and others},\n  year={2024},\n  howpublished={\\url{https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub}},\n}\n```\n\n## Acknowledgements\n\nThank you to all of our wonderful contributors!\n\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fgraphs\u002Fcontributors\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fkeras-team_keras-hub_readme_127d0c881eaf.png\" \u002F>\n\u003C\u002Fa>","# KerasHub：多框架预训练模型\n[![](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fworkflows\u002FTests\u002Fbadge.svg?branch=master)](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Factions?query=workflow%3ATests+branch%3Amaster)\n![Python](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-v3.11.0+-success.svg)\n[![Kaggle Models](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FKaggle-Models-brightgreen?colorA=0099ff)](https:\u002F\u002Fwww.kaggle.com\u002Forganizations\u002Fkeras\u002Fmodels)\n[![contributions welcome](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fcontributions-welcome-brightgreen.svg?style=flat)](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues)\n\n> [!IMPORTANT]\n> 📢 KerasNLP 现已更名为 KerasHub！📢 阅读 [公告](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues\u002F1831)。\n\n**KerasHub** 是一个旨在简单、灵活且快速的预训练建模库。该库提供了流行模型架构的 [Keras 3](https:\u002F\u002Fkeras.io\u002Fkeras_3\u002F) 实现，并搭配了可在 [Kaggle Models](https:\u002F\u002Fwww.kaggle.com\u002Forganizations\u002Fkeras\u002Fmodels) 上获取的一系列预训练检查点（checkpoints）。这些模型可用于文本、图像和音频数据，执行生成、分类及其他许多内置任务。\n\nKerasHub 是核心 Keras API 的扩展；KerasHub 组件以 `Layer`（层）和 `Model`（模型）的实现形式提供。如果你熟悉 Keras，恭喜你！你已经理解了 KerasHub 的大部分内容。\n\n所有模型都支持从单个模型定义中运行 JAX、TensorFlow 和 PyTorch，并且开箱即用即可在 GPU 和 TPU 上进行微调。模型可以使用内置的 PEFT（参数高效微调）技术在单个加速器上进行训练，或使用模型和数据并行训练进行大规模微调。请参阅我们的 [入门指南](https:\u002F\u002Fkeras.io\u002Fguides\u002Fkeras_hub\u002Fgetting_started) 开始学习我们的 API。\n\n## 快速链接\n\n### 面向所有人\n\n- [主页](https:\u002F\u002Fkeras.io\u002Fkeras_hub)\n- [入门指南](https:\u002F\u002Fkeras.io\u002Fkeras_hub\u002Fgetting_started)\n- [教程](https:\u002F\u002Fkeras.io\u002Fkeras_hub\u002Fguides)\n- [API 文档](https:\u002F\u002Fkeras.io\u002Fkeras_hub\u002Fapi)\n- [预训练模型](https:\u002F\u002Fkeras.io\u002Fkeras_hub\u002Fpresets\u002F)\n\n### 面向贡献者\n\n- [贡献征集](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues\u002F1835)\n- [路线图](https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues\u002F1836)\n- [贡献指南](CONTRIBUTING.md)\n- [风格指南](STYLE_GUIDE.md)\n- [API 设计指南](API_DESIGN_GUIDE.md)\n\n## 快速开始\n\n选择一个后端：\n\n```python\nimport os\nos.environ[\"KERAS_BACKEND\"] = \"jax\"  # 或者 \"tensorflow\" 或 \"torch\"!\n```\n\n导入 KerasHub 和其他库：\n\n```python\nimport keras\nimport keras_hub\nimport numpy as np\nimport tensorflow_datasets as tfds\n```\n\n加载一个 ResNet 模型并使用它来预测图像的标签：\n\n```python\nclassifier = keras_hub.models.ImageClassifier.from_preset(\n    \"resnet_50_imagenet\",\n    activation=\"softmax\",\n)\nurl = \"https:\u002F\u002Fupload.wikimedia.org\u002Fwikipedia\u002Fcommons\u002Fa\u002Faa\u002FCalifornia_quail.jpg\"\npath = keras.utils.get_file(origin=url)\nimage = keras.utils.load_img(path)\npreds = classifier.predict(np.array([image]))\nprint(keras_hub.utils.decode_imagenet_predictions(preds))\n```\n\n加载一个 BERT 模型并在 IMDb 电影评论上进行微调：\n\n```python\nclassifier = keras_hub.models.TextClassifier.from_preset(\n    \"bert_base_en_uncased\",\n    activation=\"softmax\",\n    num_classes=2,\n)\nimdb_train, imdb_test = tfds.load(\n    \"imdb_reviews\",\n    split=[\"train\", \"test\"],\n    as_supervised=True,\n    batch_size=16,\n)\nclassifier.fit(imdb_train, validation_data=imdb_test)\npreds = classifier.predict([\"What an amazing movie!\", \"A total waste of time.\"])\nprint(preds)\n```\n\n## 安装\n\n要安装最新的 KerasHub 版本（配合 Keras 3），只需运行：\n\n```\npip install --upgrade keras-hub\n```\n\n要安装 KerasHub 和 Keras 的最新 nightly（每日构建）更改，你可以使用我们的 nightly 包。\n\n```\npip install --upgrade keras-hub-nightly\n```\n\n目前，安装 KerasHub 将始终拉取 TensorFlow 以供 `tf.data` API 进行预处理使用。当使用 `tf.data` 进行预处理时，训练仍可在任何后端上进行。\n\n访问 [核心 Keras 入门页面](https:\u002F\u002Fkeras.io\u002Fgetting_started\u002F) 了解有关安装 Keras 3、加速器支持和与不同框架兼容性的更多信息。\n\n## 配置你的后端\n\n如果你的环境中已安装 Keras 3（见上文安装部分），你可以使用 KerasHub 配合 JAX、TensorFlow 和 PyTorch 中的任何一个。为此，请设置 `KERAS_BACKEND` 环境变量。例如：\n\n```shell\nexport KERAS_BACKEND=jax\n```\n\n或在 Colab 中，使用：\n\n```python\nimport os\nos.environ[\"KERAS_BACKEND\"] = \"jax\"\n\nimport keras_hub\n```\n\n> [!IMPORTANT]\n> 确保在导入任何 Keras 库**之前**设置 `KERAS_BACKEND`；它将在首次导入时用于设置 Keras。\n\n## 兼容性\n\n我们遵循 [语义化版本控制（Semantic Versioning）](https:\u002F\u002Fsemver.org\u002F)，并计划为使用我们组件构建的代码和保存模型提供向后兼容性保证。在我们继续进行预发布 `0.y.z` 开发期间，我们可能会随时破坏兼容性，API 不应被视为稳定。\n\n## 免责声明\n\nKerasHub 通过 `keras_hub.models` API 提供对预训练模型的访问。这些预训练模型按“原样”提供，没有任何形式的担保或条件。以下底层模型由第三方提供，并受单独许可约束：BART, BLOOM, DeBERTa, DistilBERT, GPT-2, Llama, Mistral, OPT, RoBERTa, Whisper, 和 XLM-RoBERTa。\n\n## 引用 KerasHub\n\n如果 KerasHub 有助于你的研究，我们感谢你的引用。这里是 BibTeX 条目：\n\n```bibtex\n@misc{kerashub2024,\n  title={KerasHub},\n  author={Watson, Matthew, and Chollet, Fran\\c{c}ois and Sreepathihalli,\n  Divyashree, and Saadat, Samaneh and Sampath, Ramesh, and Rasskin, Gabriel and\n  and Zhu, Scott and Singh, Varun and Wood, Luke and Tan, Zhenyu and Stenbit,\n  Ian and Qian, Chen, and Bischof, Jonathan and others},\n  year={2024},\n  howpublished={\\url{https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub}},\n}\n```\n\n## 致谢\n\n感谢所有出色的贡献者！\n\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fgraphs\u002Fcontributors\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fkeras-team_keras-hub_readme_127d0c881eaf.png\" \u002F>\n\u003C\u002Fa>","# KerasHub 快速上手指南\n\n**KerasHub** 是一个简单、灵活且快速的预训练模型库，旨在为文本、图像和音频数据提供生成、分类等多种任务的内置解决方案。它基于 [Keras 3](https:\u002F\u002Fkeras.io\u002Fkeras_3\u002F) 构建，支持 JAX、TensorFlow 和 PyTorch 三种后端，并可直接在 Kaggle Models 上获取预训练权重。\n\n## 环境准备\n\n- **Python 版本**: 建议 Python 3.11.0 及以上。\n- **核心依赖**: 需预先安装 Keras 3。\n- **后端配置**: 支持 JAX、TensorFlow 或 PyTorch。使用前需通过环境变量指定后端（见下文）。\n- **数据处理**: 示例中使用了 `tensorflow_datasets`，运行完整示例前建议安装。\n\n## 安装步骤\n\n使用 pip 安装最新版本的 KerasHub：\n\n```bash\npip install --upgrade keras-hub\n```\n\n如需体验最新的夜间构建版本（Nightly Build），可运行：\n\n```bash\npip install --upgrade keras-hub-nightly\n```\n\n> **注意**：安装 KerasHub 时会自动引入 TensorFlow 以支持 `tf.data` 预处理 API，但这不影响您在其他后端上进行训练。\n\n## 基本使用\n\n### 1. 配置后端\n\n在使用任何 Keras 库之前，必须先设置 `KERAS_BACKEND` 环境变量。例如选择 JAX 后端：\n\n```python\nimport os\nos.environ[\"KERAS_BACKEND\"] = \"jax\"  # 可选 \"tensorflow\" 或 \"torch\"\n```\n\n### 2. 导入库\n\n```python\nimport keras\nimport keras_hub\nimport numpy as np\nimport tensorflow_datasets as tfds\n```\n\n### 3. 图像分类示例\n\n加载 ResNet 模型并对图片进行预测：\n\n```python\nclassifier = keras_hub.models.ImageClassifier.from_preset(\n    \"resnet_50_imagenet\",\n    activation=\"softmax\",\n)\nurl = \"https:\u002F\u002Fupload.wikimedia.org\u002Fwikipedia\u002Fcommons\u002Fa\u002Faa\u002FCalifornia_quail.jpg\"\npath = keras.utils.get_file(origin=url)\nimage = keras.utils.load_img(path)\npreds = classifier.predict(np.array([image]))\nprint(keras_hub.utils.decode_imagenet_predictions(preds))\n```\n\n### 4. 文本分类示例\n\n加载 BERT 模型并在 IMDb 电影评论数据集上进行微调：\n\n```python\nclassifier = keras_hub.models.TextClassifier.from_preset(\n    \"bert_base_en_uncased\",\n    activation=\"softmax\",\n    num_classes=2,\n)\nimdb_train, imdb_test = tfds.load(\n    \"imdb_reviews\",\n    split=[\"train\", \"test\"],\n    as_supervised=True,\n    batch_size=16,\n)\nclassifier.fit(imdb_train, validation_data=imdb_test)\npreds = classifier.predict([\"What an amazing movie!\", \"A total waste of time.\"])\nprint(preds)\n```\n\n---\n\n更多详细文档、API 参考及预训练模型列表，请访问 [KerasHub 官方主页](https:\u002F\u002Fkeras.io\u002Fkeras_hub)。","面对紧迫的项目排期，某电商公司数据团队需要在两周内上线新品类图片自动分类功能，以优化用户搜索体验并减少人工审核成本。\n\n### 没有 keras-hub 时\n- 工程师需从零搭建网络结构或手动处理复杂的模型权重文件下载、校验与加载逻辑，极易出错。\n- 若想尝试不同硬件加速方案，必须为 TensorFlow、PyTorch 分别维护多套代码实现，迁移成本极高。\n- 缺乏标准预训练模型库，导致重复造轮子，且无法直接利用 Kaggle 上的优质开源资源。\n- 数据预处理与模型推理之间的衔接繁琐，自定义管道容易引入格式错误影响最终预测精度。\n\n### 使用 keras-hub 后\n- 通过 `from_preset` 一键加载 ResNet 等主流架构，省去了底层模型构建的繁琐步骤与依赖管理。\n- 单一定义即可兼容 JAX、TensorFlow 和 PyTorch 后端，轻松切换部署环境而不改核心业务代码。\n- 直接集成 Kaggle 预训练权重，结合内置 PEFT 技术快速完成特定品类数据的微调与迭代。\n- 提供统一的输入输出接口与图像解码工具，确保数据处理流程稳定且高效，降低运维难度。\n\nkeras-hub 通过标准化预训练模型接入流程，让团队能专注于业务逻辑而非底层基建，显著缩短了从原型验证到生产部署的时间。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fkeras-team_keras-hub_27dba404.png","keras-team","Keras","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fkeras-team_dd76ba2a.jpg","Deep Learning for humans",null,"keras-users@googlegroups.com","https:\u002F\u002Fkeras.io\u002F","https:\u002F\u002Fgithub.com\u002Fkeras-team",[84,88,92],{"name":85,"color":86,"percentage":87},"Python","#3572A5",100,{"name":89,"color":90,"percentage":91},"Shell","#89e051",0,{"name":93,"color":94,"percentage":91},"Dockerfile","#384d54",972,331,"2026-04-02T21:28:27","Apache-2.0","未说明",{"notes":101,"python":102,"dependencies":103},"必须在使用任何 Keras 库之前设置 KERAS_BACKEND 环境变量（可选 jax\u002Ftensorflow\u002Ftorch）；安装时会自动包含 TensorFlow 用于数据预处理；支持 GPU 和 TPU 微调；部分底层模型受第三方许可证约束。","3.11.0+",[104,67,105,106,107],"keras>=3.0","numpy","tensorflow","tensorflow-datasets",[26,14,13],[110,111,112,113,106,114,115,116,117,118,119],"deep-learning","keras","machine-learning","nlp","jax","llm","natural-language-processing","python","pytorch","cv","2026-03-27T02:49:30.150509","2026-04-06T05:44:27.010083",[123,128,132,137,141,146],{"id":124,"question_zh":125,"answer_zh":126,"source_url":127},2293,"KerasNLP 在 Kaggle 平台上的 Python 版本支持情况如何？","Kaggle 已将 Python 版本升级至 3.10，因此不再需要使用之前针对旧版本（如 3.7）的变通方法（workaround）来运行 KerasNLP。","https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues\u002F726",{"id":129,"question_zh":130,"answer_zh":131,"source_url":127},2294,"如何在 Kaggle 上更方便地预装或使用 KerasNLP？","社区建议向 Kaggle 提交 PR，在其 Docker 镜像的 requirements 文件中添加 `pip install keras-nlp`，以便实现预安装，减少手动配置步骤。",{"id":133,"question_zh":134,"answer_zh":135,"source_url":136},2295,"KerasNLP 中是否提供了 Mistral 模型的预设（preset）和文档？","目前尚未找到 Mistral 的预设及文档。由于 Mistral 是多语言模型，其 tokenizer 应支持多语言而非仅英语，建议关注后续更新。","https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues\u002F1418",{"id":138,"question_zh":139,"answer_zh":140,"source_url":136},2296,"将大型模型转换为 TFLite 时遇到 `Message tensorflow.GraphDef exceeds maximum protobuf size of 2GB` 错误怎么办？","这是由 TensorFlow 核心本身的 protobuf 限制导致的。目前可能无法导出超过 2GB 的模型，需等待后续修复或寻找替代转换方案。",{"id":142,"question_zh":143,"answer_zh":144,"source_url":145},2297,"在哪里可以找到 KerasNLP 的官方示例教程？","可以在 keras.io 网站上查找 KerasNLP 相关的示例。如果有新的示例想法，可以通过 GitHub Issue 进行讨论，然后拆分出具体任务进行贡献。","https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues\u002F189",{"id":147,"question_zh":148,"answer_zh":149,"source_url":150},2298,"如何为 Keras.io 贡献新的 KerasNLP 示例代码？","建议直接在 keras.io 上打开 PR，维护者会在上面进行逐行评论审查。也可以关注仓库中标记为 \"contributions welcome\" 的问题来获取入门机会。","https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fissues\u002F754",[152,157,162,167,172,177,182,187,192,197,202,207,212,217,222,227,232,237,242,247],{"id":153,"version":154,"summary_zh":155,"released_at":156},101848,"v0.27.1","## New Models\r\n- **Gemma 4**: Multimodal architecture supporting Audio, Text, and Vision, featuring both Dense and Mixture-of-Experts (MoE) variants.\r\n-  **T5Gemma 2**: Integration of the T5Gemma2 model and presets to the Hub.\r\n- **Embedding Gemma 3**: Presets and conversion scripts to load weights from HuggingFace to KerasHub.\r\n- **VideoPrism**: A powerful video understanding model using a factorized encoder design for spatial and temporal processing, supporting both video-only and multimodal vision-language tasks.\r\n- **MetaCLIP 2**: Open-source implementation of CLIP by Meta.\r\n\r\n\r\n## New Features\r\n- **TF-free Tokenizers**: Pure Python\u002FKeras implementations of `SentencePieceTokenizer`, `BytePairTokenizer`, `StartEndPacker`, and `MultiSegmentPacker`, removing TensorFlow dependencies for lighter deployments.\r\n\r\n## Export to Safetensors\r\n- **Sequential Streaming Optimization**: Optimized PyTorch memory overhead during Safetensors export via sequential streaming.\r\n- **GPT-2 Safetensors**: Migrated GPT-2 checkpoints from H5 to Hugging Face SafeTensors format.\r\n\r\n## Bug Fixes and Improvements\r\n- **VGG Architecture**: Fixed VGG architecture and updated checkpoints with activation fixes in FC layers.\r\n- **OpenVino Backend**: Fixed text generation (`generate`) for the OpenVino backend.\r\n\r\n\r\n## Contributors\r\nWe would like to thank our contributors for this release: [@bermeitinger-b](https:\u002F\u002Fgithub.com\u002Fbermeitinger-b), [@divyashreepathihalli](https:\u002F\u002Fgithub.com\u002Fdivyashreepathihalli), [@hertschuh](https:\u002F\u002Fgithub.com\u002Fhertschuh), [@james77777778](https:\u002F\u002Fgithub.com\u002Fjames77777778), [@LakshmiKalaKadali](https:\u002F\u002Fgithub.com\u002FLakshmiKalaKadali), [@laxmareddyp](https:\u002F\u002Fgithub.com\u002Flaxmareddyp), [@sachinprasadhs](https:\u002F\u002Fgithub.com\u002Fsachinprasadhs), [@sineeli](https:\u002F\u002Fgithub.com\u002Fsineeli).\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.26.0...v0.27.1","2026-04-03T01:19:02",{"id":158,"version":159,"summary_zh":160,"released_at":161},101849,"v0.27.0","## New Models\r\n- **Gemma 4**: Multimodal architecture supporting Audio, Text, and Vision, featuring both Dense and Mixture-of-Experts (MoE) variants.\r\n-  **T5Gemma 2**: Integration of the T5Gemma2 model and presets to the Hub.\r\n- **Embedding Gemma 3**: Presets and conversion scripts to load weights from HuggingFace to KerasHub.\r\n- **VideoPrism**: A powerful video understanding model using a factorized encoder design for spatial and temporal processing, supporting both video-only and multimodal vision-language tasks.\r\n- **MetaCLIP 2**: Open-source implementation of CLIP by Meta.\r\n\r\n\r\n## New Features\r\n- **TF-free Tokenizers**: Pure Python\u002FKeras implementations of `SentencePieceTokenizer`, `BytePairTokenizer`, `StartEndPacker`, and `MultiSegmentPacker`, removing TensorFlow dependencies for lighter deployments.\r\n\r\n## Export to Safetensors\r\n- **Sequential Streaming Optimization**: Optimized PyTorch memory overhead during Safetensors export via sequential streaming.\r\n- **GPT-2 Safetensors**: Migrated GPT-2 checkpoints from H5 to Hugging Face SafeTensors format.\r\n\r\n## Bug Fixes and Improvements\r\n- **VGG Architecture**: Fixed VGG architecture and updated checkpoints with activation fixes in FC layers.\r\n- **OpenVino Backend**: Fixed text generation (`generate`) for the OpenVino backend.\r\n\r\n\r\n## Contributors\r\nWe would like to thank our contributors for this release: [@bermeitinger-b](https:\u002F\u002Fgithub.com\u002Fbermeitinger-b), [@divyashreepathihalli](https:\u002F\u002Fgithub.com\u002Fdivyashreepathihalli), [@hertschuh](https:\u002F\u002Fgithub.com\u002Fhertschuh), [@james77777778](https:\u002F\u002Fgithub.com\u002Fjames77777778), [@LakshmiKalaKadali](https:\u002F\u002Fgithub.com\u002FLakshmiKalaKadali), [@laxmareddyp](https:\u002F\u002Fgithub.com\u002Flaxmareddyp), [@sachinprasadhs](https:\u002F\u002Fgithub.com\u002Fsachinprasadhs), [@sineeli](https:\u002F\u002Fgithub.com\u002Fsineeli).\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.26.0...v0.27.0\r\n","2026-04-02T23:02:27",{"id":163,"version":164,"summary_zh":165,"released_at":166},101850,"v0.26.0","## New Models\r\n\r\n\r\n*   **Translate Gemma**:  A multimodal variant of the Gemma 3 model fine-tuned for high-quality machine translation across 55 languages and capable to translate many more languages both directions, supporting both text and image inputs.\r\n*   **SAM3 (Segment Anything Model 3)**: A next-generation computer vision model that introduces Promptable Concept Segmentation (PCS), allowing for precise object and concept segmentation through text or visual prompts.\r\n*   **Qwen 2.5 Coder**: A code-specialized version of the Qwen 2.5 series, optimized for programming tasks, debugging, and code generation across a wide variety of programming languages.\r\n*   **Qwen 2.5 Math**: A specialized variant of the Qwen 2.5 family designed for advanced mathematical reasoning, capable of solving complex problems with high precision.\r\n*   **Qwen 3 Coder**: An advanced coding Mixture-of-Experts model built on the Qwen 3 MoE architecture, delivering exceptional performance across both programming benchmarks and agentic tasks.\r\n*   **RWKV7**: A high-performance, fully recurrent (100% RNN) architecture featuring linear-time complexity and constant-space inference. By eliminating the need for a KV-cache and standard attention mechanisms.\r\n\r\n## Export to Safetensors\r\n* Added Gemma3 Text models support for Safetensor export.\r\n* Added Qwen Text models support for Safetensor export.\r\n\r\n## New Features\r\n* **Hugging Face Porting Script**: Added an automated script to port any text-only decoder LLM from Hugging Face to the Keras Hub repository.\r\n* **AWQ Support**: Added support for Activation-aware Weight Quantization (AWQ).\r\n\r\n## Bug Fixes and Improvements\r\n* **Python 3.13 Compatibility**: Made `tensorflow-text` an optional dependency to ensure compatibility with Python 3.13.\r\n* **Masking**: Fixed masking issues in `TokenAndPositionEmbedding` and improved compatibility with JAX.\r\n* **Security**: Fixed a safe mode bypass vulnerability in tokenizers.\r\n* **Numerical Stability**: Fixed a float16 overflow issue in Gemma 3.\r\n\r\n## Contributors\r\nWe would like to thank our contributors for this release:\r\n@Amitavoo, @amitsrivastava78, @divyashreepathihalli, @gaurides, @hertschuh, @james77777778, @jaytiwarihub, @JyotinderSingh, @kharshith-k, @LakshmiKalaKadali, @laxmareddyp, @mattdangerw, @nikolasavic3, @pass-lin, @pctablet505, @sachinprasadhs, @shashaka.\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.25.1...v0.26.0","2026-02-06T06:55:53",{"id":168,"version":169,"summary_zh":170,"released_at":171},101851,"v0.26.0.dev0","## What's Changed\r\n* Fix reversible embedding quantization by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2476\r\n* Add FunctionGemma checkpoints to kerasHub by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2480\r\n* Models should reference ReversibleEmbedding from Keras core by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2482\r\n* Revert \"Models should reference ReversibleEmbedding from Keras core (… by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2491\r\n* Revert \"Fix reversible embedding quantization\" by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2487\r\n* Fix caching in workflows. by @hertschuh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2488\r\n* update future dev verison by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2485\r\n* update python version by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2489\r\n* An automated script to port any Text-only decoder LLM model from Hugging Face to Keras Hub repo by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2497\r\n* Add test file for convert_gpt_oss.py by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2499\r\n* Model Export to liteRT by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2405\r\n* Fix broken MiTBackbone example in SegFormerBackbone docs by @shashaka in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2503\r\n* Gemma3 text keras hf checkpoint conversion by @kharshith-k in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2433\r\n* Use `subprocess.run` in `pip_build.py` to escape wheel path. by @hertschuh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2509\r\n* Fix masking in `TokenAndPositionEmbedding` and with JAX. by @hertschuh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2510\r\n* Add rqvae model by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2490\r\n* [Bugfix][Gemma3] Check if the input has image(s) before any image processing by @gaurides in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2508\r\n* ADD RWKV7 by @pass-lin in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2421\r\n* update non_max_supression.py by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2506\r\n* Qwen keras model to HF safetensor format by @LakshmiKalaKadali in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2516\r\n* Fix safe mode bypass vulnerability in tokenizers by @amitsrivastava78 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2517\r\n* Adds support for AWQ to use `get_quantization_layer_structure` hooks by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2511\r\n* Fix overflow issue in Gemma3 float16 by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2519\r\n* Update testcase.py  by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2512\r\n* Add EDRec by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2514\r\n* Enable newly released Med Gemma 1.5  4B  variant version to hub by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2524\r\n* Remove keras-hub[nlp] note by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2531\r\n* Doc: Fix parameter name typo in BertBackbone docstring by @jaytiwarihub in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2529\r\n* skip test temporarily by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2539\r\n* Downgrading transformers version to maintain compatibility with codebase by @kharshith-k in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2542\r\n* Map new HF MedGemma model presets to conversion script  by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2541\r\n* Fixes Flux model LiteRT export failures by reducing tensor dimensions and optimizing test model size. by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2515\r\n* Add SAM3 Promptable Concept Segmentation (PCS) model by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2534\r\n* Register Qwen2.5-Coder presets by @Amitavoo in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2547\r\n* fix perset convert by @pass-lin in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2556\r\n* Fix exception handling for file errors by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2565\r\n* Remove line break from print function call by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2564\r\n* Update Kaggle Models link to Keras organization page by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2560\r\n* Perf: Pass None for missing images in Gemma3 to resolve TODO by @jaytiwarihub in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2535\r\n* Add LiteRT support for SAM3 by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2563\r\n* Add Qwen 2.5 Math model presets and checkpoint conversion support by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2566\r\n* Temporary skip for failing litert test by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2568\r\n* Add ","2026-02-06T00:54:55",{"id":173,"version":174,"summary_zh":175,"released_at":176},101852,"v0.25.1","## What's Changed\r\n\r\n* Fix float16 overflow in Gemma3 by addressing precision-related instabilities.\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.25.0...v0.25.1","2026-01-13T21:18:22",{"id":178,"version":179,"summary_zh":180,"released_at":181},101853,"v0.25.0","## Summary:\r\n\r\n### New Models:\r\nWe've integrated new open-weight models to expand the capabilities of KerasHub, featuring specialized tools for function calling and safety, as well as high-performance open-source reasoning models:\r\n\r\n*   **FunctionGemma:** We have added support for `FunctionGemma`, a lightweight model from Google built on the `Gemma 3` `270M` architecture. Designed specifically for text-only function calling, this model is optimized for single-turn scenarios and deployment in resource-constrained environments.\r\n*   **GPT OSS:** We have integrated OpenAI’s `gpt-oss` family, including the `20B` and `120B` parameter variants. These models utilize a Mixture-of-Experts (MoE) architecture with a `128k` token context window, optimized for STEM, coding, and general reasoning tasks.\r\n*   **GPT OSS Safeguard:** A new open-weight safety reasoning model from OpenAI. Built upon the `GPT OSS` architecture, it enables adaptable content classification and input-output filtering based on custom safety policies.\r\n\r\n## What's Changed\r\n* Add Gemma3 Conversion script to port weights from HF directly by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2445\r\n* Fix ESM attention for TFLite compatibility by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2466\r\n* Fix PARSeq decoder for TFLite compatibility by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2467\r\n* Fix SAM tests to make it work with Keras master by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2469\r\n* Generated GPT_OSS model files through porter script. by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2384\r\n* Register OpenAI GPT-OSS and GPT-OSS-SAFEGUARD Presets to kerashub. by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2473\r\n* Update latest synced models by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2475\r\n* Version bump to 0.25.0.dev0 by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2478\r\n* Version bump 0.25.0 by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2481\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.24.0...v0.25.0","2025-12-18T16:33:18",{"id":183,"version":184,"summary_zh":185,"released_at":186},101854,"v0.25.0.dev0","## What's Changed\r\n* Add Gemma3 Conversion script to port weights from HF directly by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2445\r\n* Fix ESM attention for TFLite compatibility by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2466\r\n* Fix PARSeq decoder for TFLite compatibility by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2467\r\n* Fix SAM tests to make it work with Keras master by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2469\r\n* Generated GPT_OSS model files through porter script. by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2384\r\n* Register OpenAI GPT-OSS and GPT-OSS-SAFEGUARD Presets to kerashub. by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2473\r\n* Update latest synced models by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2475\r\n* Version bump to 0.25.0.dev0 by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2478\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.24.0...v0.25.0.dev0","2025-12-18T08:42:41",{"id":188,"version":189,"summary_zh":190,"released_at":191},101855,"v0.24.0","## Summary:\r\n\r\n### New Models:\r\nWe've integrated new models and presets to expand the capabilities of KerasHub:\r\n\r\n*   **DINOv3:** We have added the DINOv3 model architecture and registered its corresponding presets.\r\n*   **MedGemma & MedSigLIP:** New presets have been registered for MedGemma and MedSigLIP, bringing specialized capabilities for medical domain tasks.\r\n*   **Qwen3 Embeddings:** We have registered embedding presets for the Qwen3 model family.\r\n\r\n### Improvements & Enhancements\r\nThis update includes infrastructure improvements and fixes:\r\n\r\n*   **GPTQ Quantization Hooks:** Added `get_quantization_layer_structure` hooks to facilitate GPTQ quantization support.\r\n*   **TensorFlow Compatibility:** Fixed `tensorflow-text` imports to ensure they do not break core TensorFlow functionality.\r\n*   **Gemini CLI Workflow:** Introduced a new workflow to support co-working with the Gemini CLI.\r\n\r\n## What's Changed\r\n* Fix tensorflow-text import to not break core tensorflow functionality by @nikolasavic3 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2448\r\n* Hf mirror sync for r 0.23 by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2451\r\n* Set dev version to 0.24.0.dev0 by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2447\r\n* Register MedGemma, MedSigLIP Presets to kerashub by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2450\r\n* Add DINOV3 with assistance from the Gemini CLI. by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2444\r\n* Add the workflow for co-working with the Gemini CLI. by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2453\r\n* Set JAX and Tensorflow GPU timeouts to 2.5 hours by @buildwithsuhana in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2439\r\n* mark preset test to extra large to skip GPU testing by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2458\r\n* add the default reviewer by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2460\r\n* Register Qwen3 Embedding Presets to Kerashub by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2455\r\n* Add Presets,Checkpoint conversion for SmolLM3 model by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2461\r\n* Adds get_quantization_layer_structure hooks for GPTQ by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2462\r\n* register dino v3 presets by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2463\r\n* update release version by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2465\r\n\r\n## New Contributors\r\n* @nikolasavic3 made their first contribution in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2448\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.23.0...v0.24.0","2025-12-05T05:31:16",{"id":193,"version":194,"summary_zh":195,"released_at":196},101856,"v0.24.0.dev0","## What's Changed\r\n* Fix tensorflow-text import to not break core tensorflow functionality by @nikolasavic3 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2448\r\n* Hf mirror sync for r 0.23 by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2451\r\n* Set dev version to 0.24.0.dev0 by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2447\r\n* Register MedGemma, MedSigLIP Presets to kerashub by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2450\r\n* Add DINOV3 with assistance from the Gemini CLI. by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2444\r\n* Add the workflow for co-working with the Gemini CLI. by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2453\r\n* Set JAX and Tensorflow GPU timeouts to 2.5 hours by @buildwithsuhana in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2439\r\n* mark preset test to extra large to skip GPU testing by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2458\r\n* add the default reviewer by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2460\r\n* Register Qwen3 Embedding Presets to Kerashub by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2455\r\n* Add Presets,Checkpoint conversion for SmolLM3 model by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2461\r\n* Adds get_quantization_layer_structure hooks for GPTQ by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2462\r\n* register dino v3 presets by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2463\r\n\r\n## New Contributors\r\n* @nikolasavic3 made their first contribution in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2448\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.23.0...v0.24.0.dev0","2025-12-04T21:42:34",{"id":198,"version":199,"summary_zh":200,"released_at":201},101857,"v0.23.0","## Summary:\r\n\r\n### New Models:\r\nWe've integrated a range of cutting-edge models, each designed to tackle specific challenges in their respective domains:\r\n\r\n- **Cell2Sentence:** A single-cell, biology-aware model built on the Gemma-2 architecture, designed to interpret complex biological data.\r\n\r\n- **T5Gemma:** A new encoder-decoder model, ideal for sequence-to-sequence tasks like translation and summarization.\r\n\r\n- **PARSeq:** An end-to-end, ViT-based model for scene text recognition (STR), excelling at reading text in natural images.\r\n\r\n- **D-FINE:** A high-performance, real-time object detection model.\r\n\r\n- **DepthAnythingV2:** A monocular depth estimation (MDE) model trained on a combination of synthetic labeled data and real-world unlabeled images.\r\n\r\n- **Qwen3 Moe:** The largest language model in the Qwen series, utilizing a Mixture-of-Experts (MoE) architecture for enhanced performance and efficiency.\r\n- **MobileNetV5:** A state-of-the-art vision encoder specifically designed for high-efficiency AI on edge devices.\r\n\r\n- **SmolLM3:** A compact yet powerful language model excelling in reasoning, long-context understanding, and multilingual capabilities.\r\n\r\n### Improvements & Enhancements\r\n\r\nThis update also includes several key improvements to enhance the platform's stability, compatibility, and flexibility:\r\n\r\n- **`export_to_transformers`:** You can now export trainable models, tokenizers, and configurations directly into the Hugging Face Transformers format  using `export_to_transformers`. This feature is currently available for Gemma models, with support for more architectures coming soon.\r\n- **OpenVINO Backend Support:** We've integrated OpenVINO inference support, enabling optimized inference for Mistral, Gemma, and GPT-2 models.\r\n- **Bidirectional Attention Mask:** Gemma models now support a bidirectional attention mask, enabling more effective fine-tuning on tasks that require understanding the full context of a sequence.\r\n- **CLIP & SD3 Model Refactor:** The CLIP and Stable Diffusion 3 models have been refactored to improve numerical stability. Updated checkpoints are now available to ensure seamless and reliable performance.\r\n\r\n\r\n## What's Changed\r\n* Register tiny Gemma presets by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2360\r\n* Update fixed preset version for gemma3 by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2362\r\n* Add generic export_to_transformers to the base classes by @Bond099 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2346\r\n* update version file in master by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2361\r\n* add styleguide for GCA code reviews by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2366\r\n* Update styleguide.md by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2370\r\n* Add T5Gemma to KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2339\r\n* Allow passing flexible positions to positional embedding layers by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2369\r\n* Supports Loading Quantized Models with `from_preset()` by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2367\r\n* PARSeq Model  by @sineeli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2089\r\n* Add D-FINE to KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2318\r\n* Fixing dtype issue by @buildwithsuhana in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2372\r\n* quantize(...) should accept a config object by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2388\r\n* [OpenVINO backend] Adding support for OpenVINO backend & support inference for Mistral & Gemma & GPT2 by @Mohamed-Ashraf273 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2350\r\n* minor modify by @pass-lin in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2386\r\n* Add bidirectional attention mask for EmbeddingGemma by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2382\r\n* Fixes by @buildwithsuhana in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2395\r\n* Disable DINO quantisation checks by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2397\r\n* Introduce D-FINE model presets in KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2376\r\n* Introduce T5Gemma model presets in KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2373\r\n* Update CLIP presets by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2400\r\n* Fix Gemma OpenVINO tests by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2402\r\n* Adds support for gemma_270m to checkpoint converter by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2380\r\n* [internal] Reorder @pytest.mark.large decorator to fix CI by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2410\r\n* Update preset map for VGG model by @sonali-kumari1 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2411\r\n* Update preset","2025-10-21T17:34:56",{"id":203,"version":204,"summary_zh":205,"released_at":206},101858,"v0.23.0.dev0","## What's Changed\r\n* Register tiny Gemma presets by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2360\r\n* Update fixed preset version for gemma3 by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2362\r\n* Add generic export_to_transformers to the base classes by @Bond099 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2346\r\n* update version file in master by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2361\r\n* add styleguide for GCA code reviews by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2366\r\n* Update styleguide.md by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2370\r\n* Add T5Gemma to KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2339\r\n* Allow passing flexible positions to positional embedding layers by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2369\r\n* Supports Loading Quantized Models with `from_preset()` by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2367\r\n* PARSeq Model  by @sineeli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2089\r\n* Add D-FINE to KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2318\r\n* Fixing dtype issue by @buildwithsuhana in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2372\r\n* quantize(...) should accept a config object by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2388\r\n* [OpenVINO backend] Adding support for OpenVINO backend & support inference for Mistral & Gemma & GPT2 by @Mohamed-Ashraf273 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2350\r\n* minor modify by @pass-lin in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2386\r\n* Add bidirectional attention mask for EmbeddingGemma by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2382\r\n* Fixes by @buildwithsuhana in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2395\r\n* Disable DINO quantisation checks by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2397\r\n* Introduce D-FINE model presets in KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2376\r\n* Introduce T5Gemma model presets in KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2373\r\n* Update CLIP presets by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2400\r\n* Fix Gemma OpenVINO tests by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2402\r\n* Adds support for gemma_270m to checkpoint converter by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2380\r\n* [internal] Reorder @pytest.mark.large decorator to fix CI by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2410\r\n* Update preset map for VGG model by @sonali-kumari1 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2411\r\n* Update preset map for T5 model by @sonali-kumari1 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2414\r\n* Update preset map values for cspnet by @dhantule in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2416\r\n* Add DepthAnythingV2. by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2377\r\n* Add Qwen3 Moe by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2260\r\n* update hf checkpoints list by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2381\r\n* Patch conversion script qwen3 moe by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2425\r\n* update SD3 & 3.5 presets by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2417\r\n* Add and Register the Qwen3_MoE Presets to Hub by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2429\r\n* Add MobileNetV5 to KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2399\r\n* For sharded weights let's not delete explicitly by @amitsrivastava78 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2431\r\n* Update Keras min Test version to 3.9 by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2434\r\n* Overrides `_post_quantize` to reset `generate_function` graph after quantization by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2436\r\n* Handles incompatible quantization mode for ReversibleEmbedding by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2435\r\n* extend PR stale and closure time by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2437\r\n* register depth anything presets by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2420\r\n* [SmolLM3] Add Backbone, CausalLM + Converter for HuggingFace Weights by @DavidLandup0 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2327\r\n* Register Cell2Sentence Presets by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2442\r\n* register parseq preset by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2438\r\n* register mobilenet presets by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2443\r\n\r\n## New Contributors\r\n* @buildwithsuhana made their first contribution in http","2025-10-20T23:35:35",{"id":208,"version":209,"summary_zh":210,"released_at":211},101859,"v0.22.2","**New Model: VaultGemma**\r\n\r\nVaultGemma is a 1-billion-parameter, 26-layer, text-only decoder model trained with sequence-level differential privacy (DP). \r\nDerived from Gemma 2, its architecture notably drops the norms after the Attention and MLP blocks and uses full attention for all layers, rather than alternating with local sliding attention. \r\nThe pretrained model is available with a 1024-token sequence length.\r\n\r\n## What's Changed\r\n* Add DP research model by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2396\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.22.1...v0.22.2","2025-09-12T15:31:20",{"id":213,"version":214,"summary_zh":215,"released_at":216},101860,"v0.22.1","## What's Changed\r\n* Patch release with Gemma3 presets fix by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2363\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.22.0...v0.22.1","2025-08-15T18:59:36",{"id":218,"version":219,"summary_zh":220,"released_at":221},101861,"v0.22.0","## Summary:\r\n\r\n### New Models:\r\nWe've integrated a range of cutting-edge models, each designed to tackle specific challenges in their respective domains:\r\n\r\n- **Gemma 3 270M:** Released Gemma 3 270M parameter model and instruction tuned, 18-layer, text-only model designed for\r\nhyper-efficient AI, particularly for task-specific fine-tuning.\r\n\r\n- **Qwen3:** A powerful, large-scale multilingual language model, excelling in various natural language processing tasks, from text generation to complex reasoning.\r\n\r\n- **DeiT:** Data-efficient Image Transformers (DeiT), specifically designed to train Vision Transformers effectively with less data, making high-performance visual models more accessible.\r\n\r\n- **HGNetV2:** An advanced version of the Hybrid-Grouped Network, known for its efficient architecture in computer vision tasks, particularly optimized for performance on diverse hardware.\r\n\r\n- **DINOV2:** A state-of-the-art Self-Supervised Vision Transformer, enabling the learning of robust visual representations without relying on explicit labels, ideal for foundation models.\r\n\r\n- **ESM & ESM2:** Evolutionary Scale Modeling (ESM & ESM2), powerful protein language models used for understanding protein sequences and structures, with ESM2 offering improved capabilities for bioinformatics research.\r\n\r\n###  Improvements & Enhancements\r\n\r\nThis update also includes several key improvements to enhance the platform's stability, compatibility, and flexibility:\r\n\r\n* Python 3.10 Minimum Support: Updated the minimum supported Python version to 3.10, ensuring compatibility with the latest libraries and features.\r\n* Gemma Conversion (Keras to SafeTensors): Added a new conversion script to effortlessly convert Gemma models from Keras format to Hugging Face's Safetensor format.\r\n* Gemma3 Conversion Script: Added conversion script for Gemma3 models, streamlining their integration into the Hugging Face ecosystem.\r\n* ViT Non-Square Image Support: Enhanced the Vision Transformer (ViT) model to now accept non-square images as input, providing greater flexibility for various computer vision applications.\r\n* LLM Left Padding Method: Added support for left padding in our LLM padding methods, offering more control and compatibility for specific model architectures and inference requirements.\r\n\r\n\r\n## What's Changed\r\nComplete list of all the changes included in this release.\r\n* register presets by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2268\r\n* Fix batch preprocessing bug in Moonshine generation by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2266\r\n* fix get_lora_target_names function by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2167\r\n* implement of leftpadding by @pass-lin in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2242\r\n* make vit compatible with non square images by @sineeli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2255\r\n* Bump up master version to 0.22.0.dev0 by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2277\r\n* Fix keras-io integration test by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2280\r\n* Add Qwen3 by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2249\r\n* Add DeiT Model by @Sohaib-Ahmed21 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2203\r\n* [HOTFIX] Add Docstring for QwenCausalLM by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2279\r\n* Fix: Correct coverage tracking for keras_hub by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2283\r\n* Update the sharded version number for Llama3 variants by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2294\r\n* Support None for max_shard_size by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2261\r\n* Sharded weights type error by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2296\r\n* Fix PaliGemmaCausalLM example. by @hertschuh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2302\r\n* Routine HF sync by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2303\r\n* Incorrect condition on sliding_window_size by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2289\r\n* Bump the python group with 2 updates by @dependabot[bot] in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2282\r\n* Modify TransformerEncoder masking documentation by @sonali-kumari1 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2297\r\n* Fix Gemma3InterleaveEmbeddings JAX inference error by ensuring indices are int32 by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2305\r\n* Update preset versions for Mixtral,Qwen-MoE and Mistral models by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2307\r\n* Fix Mistral conversion script by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2306\r\n* Bump the python group with 6 updates by @dependabot[bot] in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2317\r\n* Qwen3 causal lm by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkera","2025-08-14T18:01:34",{"id":223,"version":224,"summary_zh":225,"released_at":226},101862,"v0.22.0.dev0","## What's Changed\r\n* Fix Roformer export symbol by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2199\r\n* Bump up master version to 0.21 by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2204\r\n* reenable test by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2188\r\n* Add xception model by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2179\r\n* Make image converter built by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2206\r\n* Qwen - Fix Preset Loader + Add Causal LM Test by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2193\r\n* Update Qwen conversion script by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2207\r\n* Revert \"Do not export Qwen for release\" by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2208\r\n* Fixes compute_output_shape for PaliGemmaVitEncoder and Gemma3VisionEncoderBlock by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2210\r\n* Python 3.12 fix by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2211\r\n* Small Gemma3 doc-string edits by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2214\r\n* Llama3.1 by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2132\r\n* Update gemma3_causal_lm_preprocessor.py by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2217\r\n* fix: apply `weights_only = True` by @b8zhong in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2215\r\n* Fix the keras_hub package for typecheckers and IDEs by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2222\r\n* Add utility to map COCO IDs to class names by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2219\r\n* Set GPU timeouts to 2 hours by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2226\r\n* Fix nightly by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2227\r\n* Another fix for nightly builds by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2229\r\n* Cast a few more input to tensors in SD3 by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2234\r\n* Fix up package build scripts again by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2230\r\n* Add qwen presets by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2241\r\n* script for converting retinanet weights from trochvision by @sineeli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2233\r\n* Sharded weights support by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2218\r\n* Add Qwen Moe  by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2163\r\n* Add Mixtral by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2196\r\n* Made label data optional for inference and adopted other required changes by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2183\r\n* Fix the layer names by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2247\r\n* Add new CSPNet preset and add manual padding. by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2212\r\n* Update the int8 quant logic in `ReversibleEmbedding` by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2250\r\n* Add Moonshine to KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2093\r\n* Add Kaggle handle for moonshine presets by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2253\r\n* Update requirements-jax-cuda.txt by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2252\r\n* Add Mixtral,Qwen-MoE presets and Update conversion script. by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2248\r\n* fix flash attention test by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2263\r\n* Fix JAX bugs for qwen moe & mixtral by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2258\r\n* Create pull_request_template.md by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2262\r\n* Update preset versions for sharded models by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2264\r\n* Add AudioToText and AudioToTextPreprocessor class stubs to enable auto class functionality by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2265\r\n* register moonshine presets by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2267\r\n* register presets by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2268\r\n* Fix batch preprocessing bug in Moonshine generation by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2266\r\n* fix get_lora_target_names function by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2167\r\n* implement of leftpadding by @pass-lin in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2242\r\n* make vit compatible with non square images by @sineeli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2255\r\n* Bump up master version to 0.22.0.dev0 by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2277\r\n* ","2025-08-13T18:32:29",{"id":228,"version":229,"summary_zh":230,"released_at":231},101863,"v0.21.1","## Summary:\r\n * Comprehensive docstrings to QwencausalLM, resolve integration test issues for Keras-IO, and  coverage tracking for Keras-Hub.\r\n## What's Changed\r\n* Add QwencausalLM docstrings, coverage tracking, keras-io integration fix by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2284\r\n* Version bump to 0.21.1 by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2285\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.21.0...v0.21.1","2025-06-03T23:28:03",{"id":233,"version":234,"summary_zh":235,"released_at":236},101864,"v0.21.0","\r\n## Summary\r\n* New Models.\r\n   * **Xception**: Added Xception architecture for image classification tasks.\r\n   * **Qwen**: Added Qwen2.5 large language models and presets of base language models and instruction-tuned language models ranging from 0.5 to 72 billion parameters.\r\n   * **Qwen MoE**: Added transformer-based Mixture of Experts (MoE) decoder-only language model with a base variant having 2.7B activated parameters during runtime.\r\n   * **Mixtral**: Added Mixtral LLM, a pretrained generative Sparse Mixture of Experts with pre-trained and instruction tuned models having 7 billion activated parameters.\r\n   * **Moonshine**: Added Moonshine, a speech recognition task model.\r\n   * **CSPNet**: Added Cross Stage Partial Network (CSPNet) classification task model.\r\n   * **Llama3**: Added support for Llama 3.1 and 3.2.\r\n \r\n* Added sharded weight support to KerasPresetSaver and KerasPresetLoader, defaulting to a 10GB maximum shard size.\r\n\r\n\r\n\r\n\r\n\r\n\r\n## What's Changed\r\n* Fix Roformer export symbol by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2199\r\n* Bump up master version to 0.21 by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2204\r\n* reenable test by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2188\r\n* Add xception model by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2179\r\n* Make image converter built by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2206\r\n* Qwen - Fix Preset Loader + Add Causal LM Test by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2193\r\n* Update Qwen conversion script by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2207\r\n* Revert \"Do not export Qwen for release\" by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2208\r\n* Fixes compute_output_shape for PaliGemmaVitEncoder and Gemma3VisionEncoderBlock by @JyotinderSingh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2210\r\n* Python 3.12 fix by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2211\r\n* Small Gemma3 doc-string edits by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2214\r\n* Llama3.1 by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2132\r\n* Update gemma3_causal_lm_preprocessor.py by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2217\r\n* fix: apply `weights_only = True` by @b8zhong in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2215\r\n* Fix the keras_hub package for typecheckers and IDEs by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2222\r\n* Add utility to map COCO IDs to class names by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2219\r\n* Set GPU timeouts to 2 hours by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2226\r\n* Fix nightly by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2227\r\n* Another fix for nightly builds by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2229\r\n* Cast a few more input to tensors in SD3 by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2234\r\n* Fix up package build scripts again by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2230\r\n* Add qwen presets by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2241\r\n* script for converting retinanet weights from trochvision by @sineeli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2233\r\n* Sharded weights support by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2218\r\n* Add Qwen Moe  by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2163\r\n* Add Mixtral by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2196\r\n* Made label data optional for inference and adopted other required changes by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2183\r\n* Fix the layer names by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2247\r\n* Add new CSPNet preset and add manual padding. by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2212\r\n* Update the int8 quant logic in `ReversibleEmbedding` by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2250\r\n* Add Moonshine to KerasHub by @harshaljanjani in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2093\r\n* Add Kaggle handle for moonshine presets by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2253\r\n* Update requirements-jax-cuda.txt by @pctablet505 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2252\r\n* Add Mixtral,Qwen-MoE presets and Update conversion script. by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2248\r\n* fix flash attention test by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2263\r\n* Fix JAX bugs for qwen moe & mixtral by @kanpuriyanawab in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2258\r\n* Create pull_request_template.md by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2262\r\n* Update preset versions for sharded models by @laxmareddyp in https:\u002F\u002Fg","2025-05-28T19:07:21",{"id":238,"version":239,"summary_zh":240,"released_at":241},101865,"v0.20.0","## What's Changed\r\n* Install TF Text on non-Windows only by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2115\r\n* Add SigLIP by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2113\r\n* Fix `PaliGemmaVitEncoder` output shape by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2123\r\n* Cspnet architecture. by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2091\r\n* Update our master version to be a dev release by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2131\r\n* Add top 3 HF Presets for Mobilenet by @pkgoogle in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2105\r\n* Add SigLIP2 by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2127\r\n* update Gemma attention for TPU by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2130\r\n* Update dev version rule for nightly by @SamanehSaadat in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2139\r\n* Fix dtype bug in image converter by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2147\r\n* Add instruction in .md for manual pre-commit run by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2148\r\n* Add Qwen 2.5 by @shivance in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2088\r\n* Updated CONTRIBUTING.md (Fixes issue #2153) by @villurignanesh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2156\r\n* Update kaggle preset paths for SigLip model by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2164\r\n* Routine Kaggle HF sync by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2165\r\n* Enable LoRA target names arg by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2166\r\n* Update retinanet_presets.py by @sineeli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2157\r\n* Add Gemma3 by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2152\r\n* Add precommit to the common requirements file by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2173\r\n* Add back a format script for compat by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2174\r\n* Add a TextToImagePreprocessor base class by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2181\r\n* Bump the python group with 2 updates by @dependabot in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2185\r\n* implement of roformerv2 by @pass-lin in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2145\r\n* Move sliding window attn before FA block for Gemma by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2187\r\n* Update gating condition to include check for supporting GPUs for flash attention by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2184\r\n* Revert \"Fix dtype bug in image converter (#2147)\" by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2180\r\n* Add vision for Gemma3 by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2170\r\n* Do not export Qwen for release by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2198\r\n* Version bump to 0.20.0.dev1 by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2200\r\n* Version bump to 0.20.0 by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2202\r\n\r\n## New Contributors\r\n* @villurignanesh made their first contribution in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2156\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.19.3...v0.20.0","2025-04-03T23:48:31",{"id":243,"version":244,"summary_zh":245,"released_at":246},101866,"v0.20.0.dev1","## What's Changed\r\n* Version bump to 0.20.0.dev1 by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2200\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.20.0.dev0...v0.20.0.dev1","2025-04-03T19:11:49",{"id":248,"version":249,"summary_zh":250,"released_at":251},101867,"v0.20.0.dev0","## What's Changed\r\n* Install TF Text on non-Windows only by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2115\r\n* Add SigLIP by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2113\r\n* Fix `PaliGemmaVitEncoder` output shape by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2123\r\n* Cspnet architecture. by @sachinprasadhs in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2091\r\n* Update our master version to be a dev release by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2131\r\n* Add top 3 HF Presets for Mobilenet by @pkgoogle in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2105\r\n* Add SigLIP2 by @james77777778 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2127\r\n* update Gemma attention for TPU by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2130\r\n* Update dev version rule for nightly by @SamanehSaadat in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2139\r\n* Fix dtype bug in image converter by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2147\r\n* Add instruction in .md for manual pre-commit run by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2148\r\n* Add Qwen 2.5 by @shivance in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2088\r\n* Updated CONTRIBUTING.md (Fixes issue #2153) by @villurignanesh in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2156\r\n* Update kaggle preset paths for SigLip model by @laxmareddyp in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2164\r\n* Routine Kaggle HF sync by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2165\r\n* Enable LoRA target names arg by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2166\r\n* Update retinanet_presets.py by @sineeli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2157\r\n* Add Gemma3 by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2152\r\n* Add precommit to the common requirements file by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2173\r\n* Add back a format script for compat by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2174\r\n* Add a TextToImagePreprocessor base class by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2181\r\n* Bump the python group with 2 updates by @dependabot in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2185\r\n* implement of roformerv2 by @pass-lin in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2145\r\n* Move sliding window attn before FA block for Gemma by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2187\r\n* Update gating condition to include check for supporting GPUs for flash attention by @divyashreepathihalli in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2184\r\n* Revert \"Fix dtype bug in image converter (#2147)\" by @mattdangerw in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2180\r\n* Add vision for Gemma3 by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2170\r\n* Do not export Qwen for release by @abheesht17 in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2198\r\n\r\n## New Contributors\r\n* @villurignanesh made their first contribution in https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fpull\u002F2156\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fkeras-team\u002Fkeras-hub\u002Fcompare\u002Fv0.19.0.dev0...v0.20.0.dev0","2025-04-03T17:58:49"]