{"data":[{"slug":"deepseek-ai-FlashMLA","title_en":"FlashMLA: Optimized Attention Kernels for DeepSeek Models","title_zh":"FlashMLA: 针对 DeepSeek 模型优化的 Attention Kernels","summary_en":"FlashMLA is a library of high-performance attention kernels specifically designed to power DeepSeek-V3 and DeepSeek-V3.2 models. It provides optimized implementations for both sparse and dense attention mechanisms during prefill and decoding stages. The library supports advanced features like FP8 KV cache and is compatible with various GPU architectures including SM90 and SM100.","summary_zh":"FlashMLA 是一个高性能 Attention Kernels 库，专为驱动 DeepSeek-V3 和 DeepSeek-V3.2 模型而设计。它为预填充和解码阶段的稀疏和密集 Attention 机制提供了优化实现。该库支持 FP8 KV cache 等高级功能，并兼容包括 SM90 和 SM100 在内的多种 GPU 架构。","category":"AI","tags":["DeepSeek","Attention","CUDA","PyTorch","LLM"],"stars":12617,"score":93,"trending_score":82,"source":"github-zh-inc","repo_url":"https://github.com/deepseek-ai/FlashMLA","rank_overall":1,"rank_category":1,"updated_at":"2026-05-01T12:03:30.738Z"},{"slug":"instantdb-instant","title_en":"Instant: A Real-time Client-side Database Backend","title_zh":"Instant: 一个实时客户端数据库后端","summary_en":"Instant is a comprehensive backend solution designed to simplify app development by providing built-in auth, permissions, storage, and real-time streams. It allows developers to write relational queries directly on the client while automatically handling data fetching, offline caching, and optimistic updates. By treating database interactions as a client-side concern, the platform enables the rapid creation of multiplayer applications with minimal boilerplate code.","summary_zh":"Instant 是一个全面的后端解决方案，旨在通过提供内置的 auth、permissions、storage 和实时 streams 来简化应用开发。它允许开发者直接在客户端编写关系型查询，同时自动处理数据获取、离线缓存和乐观更新。通过将数据库交互视为客户端关注点，该平台能够以最少的样板代码快速创建多人协作应用程序。","category":"Database","tags":["PostgreSQL","Real-time","GraphQL","React","Backend"],"stars":10217,"score":92,"trending_score":1145,"source":"hn","repo_url":"https://github.com/instantdb/instant","rank_overall":2,"rank_category":1,"updated_at":"2026-05-01T12:05:54.664Z"},{"slug":"BerriAI-litellm","title_en":"LiteLLM: Unified AI Gateway for 100+ LLMs","title_zh":"LiteLLM: 适用于 100+ LLM 的统一 AI 网关","summary_en":"LiteLLM provides a unified interface to interact with over 100 LLM providers using a consistent OpenAI-compatible format. Developers can utilize it as a Python SDK for direct integration or deploy it as a production-ready proxy server. The platform simplifies LLM management by offering features like load balancing, spend tracking, and virtual keys.","summary_zh":"LiteLLM 提供了一个统一的接口，可以使用一致的 OpenAI 兼容格式与 100 多个 LLM 提供商进行交互。开发人员可以将其用作 Python SDK 进行直接集成，或者将其部署为生产就绪的代理服务器。该平台通过提供负载均衡、支出跟踪和虚拟密钥等功能，简化了 LLM 的管理流程。","category":"AI","tags":["LLM","Gateway","Proxy","API","Python"],"stars":45389,"score":89,"trending_score":938,"source":"hn","repo_url":"https://github.com/BerriAI/litellm","rank_overall":3,"rank_category":2,"updated_at":"2026-05-01T12:05:21.775Z"},{"slug":"TabbyML-tabby","title_en":"Tabby: Self-Hosted AI Coding Assistant","title_zh":"Tabby: 自托管 AI 编程助手","summary_en":"Tabby is a self-hosted, open-source AI coding assistant designed as an on-premises alternative to GitHub Copilot. It operates as a self-contained system that does not require external cloud services or database management. The platform supports consumer-grade GPUs and offers an OpenAPI interface for seamless integration with existing development infrastructure.","summary_zh":"Tabby 是一款自托管的开源 AI 编程助手，旨在作为 GitHub Copilot 的本地化替代方案。它作为一个独立的系统运行，无需外部云服务或数据库管理。该平台支持消费级 GPU，并提供 OpenAPI 接口，以便与现有的开发基础设施无缝集成。","category":"AI","tags":["Machine Learning","Self-hosted","Rust","Docker","LLM"],"stars":33480,"score":88,"trending_score":627,"source":"hn","repo_url":"https://github.com/TabbyML/tabby","rank_overall":4,"rank_category":3,"updated_at":"2026-05-01T12:05:35.697Z"},{"slug":"pingcap-tidb","title_en":"TiDB: A Distributed SQL Database","title_zh":"TiDB: 一个分布式 SQL 数据库","summary_en":"TiDB is an open-source, cloud-native, distributed SQL database designed for high availability and horizontal scalability. It supports hybrid transactional and analytical processing by utilizing separate row-based and columnar storage engines. The platform maintains strong consistency and is fully compatible with the MySQL 8.0 protocol.","summary_zh":"TiDB 是一个开源的云原生分布式 SQL 数据库，专为高可用性和水平扩展性而设计。它通过利用独立的行式和列式存储引擎，支持混合事务和分析处理。该平台保持强一致性，并与 MySQL 8.0 协议完全兼容。","category":"Database","tags":["SQL","Distributed Systems","HTAP","Cloud-Native","Go"],"stars":40048,"score":87,"trending_score":92,"source":"github-zh-inc","repo_url":"https://github.com/pingcap/tidb","rank_overall":5,"rank_category":2,"updated_at":"2026-05-01T12:03:44.632Z"},{"slug":"deepseek-ai-DeepEP","title_en":"DeepEP: High-Performance Expert Parallelism Library","title_zh":"DeepEP: 高性能专家并行库","summary_en":"DeepEP is a high-performance communication library designed for modern machine learning training and inference, specifically focusing on expert parallelism. The library utilizes a lightweight Just-In-Time compilation module and the NCCL Gin backend to deliver high-throughput, low-latency GPU kernels. It supports advanced features like pipeline parallelism and remote memory access while significantly reducing SM resource consumption compared to previous versions.","summary_zh":"DeepEP 是一款专为现代机器学习训练和推理设计的高性能通信库，特别专注于专家并行技术。该库利用轻量级的 Just-In-Time 编译模块和 NCCL Gin 后端，提供高吞吐量、低延迟的 GPU kernels。它支持流水线并行和远程内存访问等高级功能，并与之前版本相比显著降低了 SM 资源消耗。","category":"AI","tags":["Machine Learning","CUDA","NCCL","Distributed Training","GPU"],"stars":9594,"score":86,"trending_score":79,"source":"github-zh-inc","repo_url":"https://github.com/deepseek-ai/DeepEP","rank_overall":6,"rank_category":4,"updated_at":"2026-05-01T12:03:26.976Z"},{"slug":"hyperdxio-hyperdx","title_en":"HyperDX Observability Platform for ClickHouse","title_zh":"HyperDX 用于 ClickHouse 的可观测性平台","summary_en":"HyperDX is an observability platform that enables engineers to search and visualize logs, traces, and metrics on top of ClickHouse clusters. It provides a schema-agnostic interface that supports full-text search and intuitive property filtering without requiring complex query languages. The tool integrates seamlessly with OpenTelemetry to help teams monitor system health and resolve production issues efficiently.","summary_zh":"HyperDX 是一个可观测性平台，使工程师能够在 ClickHouse 集群之上搜索和可视化日志、追踪和指标。它提供了一个与模式无关的界面，支持全文搜索和直观的属性过滤，而无需复杂的查询语言。该工具与 OpenTelemetry 无缝集成，帮助团队监控系统健康状况并高效解决生产问题。","category":"DevOps","tags":["Observability","ClickHouse","OpenTelemetry","Logging","Monitoring"],"stars":9473,"score":86,"trending_score":722,"source":"hn","repo_url":"https://github.com/hyperdxio/hyperdx","rank_overall":7,"rank_category":1,"updated_at":"2026-05-01T12:05:51.360Z"},{"slug":"supabase-postgres_lsp","title_en":"Postgres Language Server","title_zh":"Postgres Language Server","summary_en":"postgres_lsp is a comprehensive toolchain and Language Server Protocol implementation designed to enhance the Postgres development experience. It utilizes the native libpg_query parser to provide 100% syntax compatibility across various editor integrations and CLI tools. The project offers a wide range of features including autocompletion, type checking, and database linting to streamline SQL workflows.","summary_zh":"postgres_lsp 是一套全面的工具链和 Language Server Protocol 实现，旨在提升 Postgres 的开发体验。它利用原生的 libpg_query 解析器，在各种编辑器集成和 CLI 工具中提供 100% 的语法兼容性。该项目提供了包括自动补全、类型检查和数据库 linting 在内的多种功能，以简化 SQL 工作流程。","category":"Database","tags":["Postgres","LSP","SQL","Developer Tools"],"stars":5225,"score":85,"trending_score":963,"source":"hn","repo_url":"https://github.com/supabase/postgres_lsp","rank_overall":8,"rank_category":3,"updated_at":"2026-05-01T12:06:17.505Z"},{"slug":"ant-design-ant-design","title_en":"Ant Design: Enterprise-Class React UI Library","title_zh":"Ant Design: 企业级 React UI 组件库","summary_en":"Ant Design is a comprehensive UI design language and React component library built for enterprise-level web applications. It provides a robust set of high-quality components written in TypeScript with support for internationalization and theme customization. The library is compatible with modern browsers, server-side rendering, and Electron environments.","summary_zh":"Ant Design 是一套专为企业级 Web 应用打造的综合性 UI 设计语言和 React 组件库。它提供了一套由 TypeScript 编写的高质量组件，并支持国际化和主题定制。该库兼容现代浏览器、服务端渲染以及 Electron 环境。","category":"Frontend","tags":["React","TypeScript","UI Library","CSS-in-JS"],"stars":97903,"score":83,"trending_score":95,"source":"github-zh-inc","repo_url":"https://github.com/ant-design/ant-design","rank_overall":9,"rank_category":1,"updated_at":"2026-05-01T12:02:54.110Z"},{"slug":"khoj-ai-khoj","title_en":"Khoj: Your Personal AI Second Brain","title_zh":"Khoj: 您的个人 AI 第二大脑","summary_en":"Khoj is a versatile personal AI application designed to extend your capabilities by integrating with various local and online large language models. It allows users to interact with their personal documents and the internet through a unified interface accessible across multiple platforms. The project is open-source and supports flexible deployment options ranging from private on-device setups to scalable enterprise cloud solutions.","summary_zh":"Khoj 是一款多功能的个人 AI 应用程序，旨在通过集成各种本地和在线大语言模型来扩展您的能力。它允许用户通过跨多个平台访问的统一界面与个人文档及互联网进行交互。该项目是开源的，并支持从私有设备端部署到可扩展的企业级云解决方案等多种灵活的部署选项。","category":"AI","tags":["LLM","Semantic Search","Self-hosted","Personal Assistant"],"stars":34330,"score":83,"trending_score":565,"source":"hn","repo_url":"https://github.com/khoj-ai/khoj","rank_overall":10,"rank_category":5,"updated_at":"2026-05-01T12:05:59.409Z"},{"slug":"Mininglamp-AI-Mano-P","title_en":"Mano-P: GUI-Aware Private AI Agent for Edge Devices","title_zh":"Mano-P: 面向边缘设备的 GUI 感知私有 AI Agent","summary_en":"Mano-P is a GUI-VLA agent project designed to enable autonomous, private task execution on edge devices like Mac mini and MacBook. It utilizes advanced reinforcement learning and edge-native inference to perform complex GUI automation, cross-system data integration, and long-task planning. The project provides a secure, local-first solution that eliminates the need for cloud API calls while maintaining high performance across various benchmarks.","summary_zh":"Mano-P 是一个 GUI-VLA Agent 项目，旨在 Mac mini 和 MacBook 等边缘设备上实现自主且私密的任务执行。它利用先进的强化学习和边缘原生推理技术，执行复杂的 GUI 自动化、跨系统数据集成以及长任务规划。该项目提供了一种安全且优先本地化的解决方案，在无需云端 API 调用且保持跨基准测试高性能的同时，确保了数据隐私。","category":"AI","tags":["Agent","Computer Vision","Edge Computing","LLM","Automation"],"stars":1264,"score":83,"trending_score":75,"source":"github-zh-em","repo_url":"https://github.com/Mininglamp-AI/Mano-P","rank_overall":11,"rank_category":6,"updated_at":"2026-05-01T12:01:42.616Z"},{"slug":"bytedance-deer-flow","title_en":"DeerFlow: An Open-Source Super Agent Harness","title_zh":"DeerFlow: 开源超级智能体框架","summary_en":"DeerFlow is an open-source super agent harness designed to orchestrate sub-agents, memory, and sandboxes for complex task execution. The platform features a ground-up rewrite in version 2.0, offering enhanced extensibility through a modular skill and tool architecture. It supports diverse deployment environments, including local development and Docker-based production setups, with integrated support for multiple messaging channels.","summary_zh":"DeerFlow 是一款开源超级智能体框架，旨在编排子智能体、记忆和沙盒以执行复杂任务。该平台在 2.0 版本中进行了彻底重构，通过模块化的技能和工具架构提供了更强的可扩展性。它支持多种部署环境，包括本地开发和基于 Docker 的生产环境，并集成了对多种消息渠道的支持。","category":"AI","tags":["LLM","Agents","Docker","Python","Automation"],"stars":64406,"score":82,"trending_score":95,"source":"github-zh-inc","repo_url":"https://github.com/bytedance/deer-flow","rank_overall":12,"rank_category":7,"updated_at":"2026-05-01T12:03:22.043Z"},{"slug":"Kanaries-pygwalker","title_en":"PyGWalker: Interactive Data Visualization for Jupyter Notebooks","title_zh":"PyGWalker: Jupyter Notebook 的交互式数据可视化工具","summary_en":"PyGWalker transforms pandas DataFrames into an interactive user interface that simplifies data analysis and visualization within Jupyter Notebooks. It integrates the Graphic Walker library to provide a drag-and-drop experience similar to Tableau for exploring and cleaning data. Users can easily create various chart types, apply filters, and perform visual data transformations directly in their existing Python workflow.","summary_zh":"PyGWalker 将 pandas DataFrames 转换为交互式用户界面，从而简化了 Jupyter Notebook 中的数据分析与可视化过程。它集成了 Graphic Walker 库，提供类似于 Tableau 的拖拽式体验，方便用户进行数据探索和清洗。用户可以直接在现有的 Python 工作流中轻松创建各种图表类型、应用筛选器并执行可视化数据转换。","category":"AI","tags":["Python","Data Visualization","Jupyter","Data Analysis"],"stars":15762,"score":82,"trending_score":712,"source":"hn","repo_url":"https://github.com/Kanaries/pygwalker","rank_overall":13,"rank_category":8,"updated_at":"2026-05-01T12:05:21.616Z"},{"slug":"mmulet-term.everything","title_en":"Term.Everything: Run GUI Applications Directly in Your Terminal","title_zh":"Term.Everything: 在终端中直接运行 GUI 应用程序","summary_en":"Term.Everything is a Linux CLI tool that functions as a Wayland compositor to render GUI applications directly within a terminal window. It supports both X11 and Wayland host systems and allows users to run complex graphical software, including web browsers and games, over SSH. The project is currently in beta and aims to enable the execution of any graphical application within a terminal environment.","summary_zh":"Term.Everything 是一个 Linux CLI 工具，它作为 Wayland 合成器在终端窗口内直接渲染 GUI 应用程序。它支持 X11 和 Wayland 主机系统，并允许用户通过 SSH 运行复杂的图形软件，包括网页浏览器和游戏。该项目目前处于测试阶段，旨在实现任何图形应用程序在终端环境中的执行。","category":"Other","tags":["Go","Wayland","CLI","Linux","Terminal"],"stars":7956,"score":82,"trending_score":1094,"source":"hn","repo_url":"https://github.com/mmulet/term.everything","rank_overall":14,"rank_category":1,"updated_at":"2026-05-01T12:06:07.847Z"},{"slug":"THUDM-slime","title_en":"slime: High-Performance LLM Post-Training Framework for RL Scaling","title_zh":"slime: 用于 RL 扩展的高性能 LLM 后训练框架","summary_en":"Slime is a specialized post-training framework designed to scale reinforcement learning for large language models. It integrates Megatron-LM for high-performance training with SGLang to provide flexible, efficient data generation workflows. The architecture decouples training and rollout processes, enabling researchers to build and deploy complex agentic RL systems.","summary_zh":"Slime 是一个专门设计的后训练框架，旨在扩展大语言模型的强化学习。它集成了 Megatron-LM 以实现高性能训练，并结合 SGLang 提供灵活且高效的数据生成工作流。该架构将训练和部署过程解耦，使研究人员能够构建并部署复杂的智能体 RL 系统。","category":"AI","tags":["LLM","Reinforcement Learning","Megatron-LM","SGLang","Post-training"],"stars":5543,"score":79,"trending_score":74,"source":"github-zh-inc","repo_url":"https://github.com/THUDM/slime","rank_overall":15,"rank_category":9,"updated_at":"2026-05-01T12:02:28.760Z"},{"slug":"PaddlePaddle-PaddleFormers","title_en":"PaddleFormers: High-Performance Large Model Training Library Based on PaddlePaddle","title_zh":"PaddleFormers：基于飞桨的高性能大模型训练库","summary_en":"PaddleFormers is a Transformers library built on the Baidu PaddlePaddle framework, designed to provide training interfaces and functional experiences for Large Language Models and Vision-Language Models equivalent to Hugging Face. By integrating tensor parallelism, pipeline parallelism, and automatic mixed precision, the project achieves training performance that surpasses Megatron-LM on mainstream models. Furthermore, it fully supports domestic computing chips and is compatible with the Safetensors format, helping developers efficiently complete the entire process from pre-training to post-training.","summary_zh":"PaddleFormers 是基于百度飞桨框架构建的 Transformers 库，旨在为大语言模型与视觉语言模型提供与 Hugging Face 对等的训练接口与功能体验。该项目通过集成张量并行、流水线并行及自动混合精度等技术，在主流模型上实现了超越 Megatron-LM 的训练性能。此外，它还全面支持国产计算芯片，并兼容 Safetensors 格式，助力开发者高效完成从预训练到后训练的全流程任务。","category":"AI","tags":["PaddlePaddle","LLM","Deep Learning","Transformers","Distributed Training"],"stars":12991,"score":78,"trending_score":82,"source":"github-zh-inc","repo_url":"https://github.com/PaddlePaddle/PaddleFormers","rank_overall":16,"rank_category":10,"updated_at":"2026-05-01T12:02:22.944Z"},{"slug":"ant-design-ant-design-pro","title_en":"Ant Design Pro: Enterprise React Boilerplate","title_zh":"Ant Design Pro: 企业级 React 脚手架","summary_en":"Ant Design Pro is an out-of-the-box UI solution designed for building enterprise-level React applications. It provides a comprehensive set of templates, blocks, and best practices to streamline the development process. The framework supports modern technologies like TypeScript, Umi Max, and Ant Design 6 to ensure high-quality, scalable code.","summary_zh":"Ant Design Pro 是一款开箱即用的 UI 解决方案，专为构建企业级 React 应用程序而设计。它提供了一套全面的模板、区块和最佳实践，旨在简化开发流程。该框架支持 TypeScript、Umi Max 和 Ant Design 6 等现代技术，以确保代码的高质量与可扩展性。","category":"Frontend","tags":["React","Ant Design","TypeScript","Tailwind CSS","UmiJS"],"stars":38149,"score":77,"trending_score":91,"source":"github-zh-inc","repo_url":"https://github.com/ant-design/ant-design-pro","rank_overall":17,"rank_category":2,"updated_at":"2026-05-01T12:03:07.453Z"},{"slug":"docmost-docmost","title_en":"Docmost: Open-Source Collaborative Wiki and Documentation Platform","title_zh":"Docmost: 开源协作 Wiki 与文档平台","summary_en":"Docmost is an open-source platform designed for collaborative wiki creation and documentation management. It offers a comprehensive suite of features including real-time editing, integrated diagramming tools, and robust permissions management. Users can organize content into spaces while utilizing advanced functionality like page history, file attachments, and multi-language support.","summary_zh":"Docmost 是一个专为协作式 Wiki 创建和文档管理而设计的开源平台。它提供了一套全面的功能，包括实时编辑、集成绘图工具以及强大的权限管理。用户可以将内容组织到不同的空间中，同时利用页面历史记录、文件附件和多语言支持等高级功能。","category":"Other","tags":["Documentation","Wiki","Collaboration","Open Source"],"stars":19988,"score":77,"trending_score":551,"source":"hn","repo_url":"https://github.com/docmost/docmost","rank_overall":18,"rank_category":2,"updated_at":"2026-05-01T12:05:45.684Z"},{"slug":"alibaba-ROLL","title_en":"ROLL: Reinforcement Learning Optimization for Large-Scale LLM Training","title_zh":"ROLL: 用于大规模 LLM 训练的强化学习优化框架","summary_en":"ROLL is an efficient, user-friendly library designed for scaling reinforcement learning workflows for large language models across large-scale GPU clusters. It supports diverse training paradigms including RLVR, agentic interaction, and distillation, while integrating advanced backends like Megatron-Core, vLLM, and SGLang. The framework provides robust observability and flexible resource management to enhance performance in complex reasoning and human preference alignment tasks.","summary_zh":"ROLL 是一个高效且易于使用的库，旨在为大规模 GPU 集群上的大语言模型扩展强化学习工作流。它支持包括 RLVR、智能体交互和蒸馏在内的多种训练范式，同时集成了 Megatron-Core、vLLM 和 SGLang 等先进后端。该框架提供强大的可观测性和灵活的资源管理，以提升复杂推理和人类偏好对齐任务中的性能。","category":"AI","tags":["Reinforcement Learning","LLM","Distributed Training","PPO","Deep Learning"],"stars":3120,"score":75,"trending_score":69,"source":"github-zh-inc","repo_url":"https://github.com/alibaba/ROLL","rank_overall":19,"rank_category":11,"updated_at":"2026-05-01T12:02:46.024Z"},{"slug":"epicenter-so-epicenter","title_en":"Epicenter: Local-First Open-Source Application Ecosystem","title_zh":"Epicenter: 本地优先的开源应用生态系统","summary_en":"Epicenter is an ecosystem of open-source, local-first applications that store data as plain text and SQLite files on your machine. It utilizes Yjs CRDTs to ensure seamless multi-device synchronization while maintaining user privacy through client-side encryption. Developers can leverage the core workspace library to build custom applications that share a unified, typed data schema.","summary_zh":"Epicenter 是一个开源的本地优先应用生态系统，将数据以纯文本和 SQLite 文件形式存储在您的设备上。它利用 Yjs CRDTs 确保多设备间的无缝同步，同时通过客户端加密保护用户隐私。开发者可以利用核心工作区库构建自定义应用程序，并共享统一的类型化数据架构。","category":"Other","tags":["Local-first","CRDT","Svelte","Tauri","TypeScript"],"stars":4491,"score":74,"trending_score":591,"source":"hn","repo_url":"https://github.com/epicenter-so/epicenter","rank_overall":20,"rank_category":3,"updated_at":"2026-05-01T12:06:04.684Z"}],"meta":{"total":93,"limit":20,"offset":0,"generated_at":"2026-08-07T12:03:32.674Z"}}