Project directory

Projects

Open-source AI projects tracked by the leaderboard, ordered by GitHub stars.

Python 2,796 stars

wenge-research/YAYI2

YAYI 2 是中科闻歌研发的新一代开源大语言模型，采用了超过 2 万亿 Tokens 的高质量、多语言语料进行预训练。(Repo for YaYi 2 Chinese LLMs)

### Overview: YAYI 2 是中科闻歌研发的**新一代开源大语言模型**，参数规模为 30B。该模型基于 Transformer 架构，采用了超过 2 万亿 Tokens 的高质量、多语言语料进行预训练。通过百万级指令微调和人类反馈强化学习，模型能够更好地与人类价值观对齐，适用于通用和特定领域的应用场景。 ### Key features: - **Model Size**: 30B - **Pre-training**: 2 万亿 Tokens 高质量多语言语料 - **Micro-tuning**: 百万级指令微调 - **Human Feedback**: 人类反馈强化学习 - **Scoring**: C-Eval、MMLU、AGIEval、CMMLU、GAOKAO-Bench、GSM8K、MATH、BBH、HumanEval、MBPP ### Tech stack: Python ### Why it matters: 通过开源 YAYI 2 模型，我们希望促进中文预训练大模型开源社区的发展，并积极为构建雅意大模型生态做出贡献。我们希望通过这一开源项目，与每一位合作伙伴共同推动雅意大模型技术的发展，从而为更多人提供更好的语言模型服务。

Java 2,795 stars

opensolon/solon

🔥 Java enterprise application development framework for full scenario: Restrained, Efficient, Open, Ecologicalll!!! 700% higher concurrency 50% memory savings Startup is 10 times faster. Packing 90% smaller; Compatible with java8 ~ java26; Supports LTS. (Replaceable spring)

Overview: Solon is a Java enterprise application development framework designed for full scenario, offering enhanced performance, productivity, and compatibility. Key features: - Better cost performance for computing resources, with 700% higher concurrency and 50% memory savings. - Faster development productivity, including less code, easy start, and 10x faster startup. - Better production and deployment experience, with a 90% smaller package. - Greater range of compatibility, including non-java-ee architecture and support for java8 ~ java26 and GraalVM native image. Tech stack: Java Why it matters: Solon is designed to provide a high-performance, efficient, and open solution for Java enterprise application development, offering significant improvements in concurrency, memory usage, and startup speed, while also supporting a wide range of compatibility options.

TeX 2,794 stars

NiuTrans/MTBook

《机器翻译：基础与模型》肖桐朱靖波著 - Machine Translation: Foundations and Models

### Overview: 《机器翻译：基础与模型》是一本旨在系统介绍机器翻译基础知识和建模方法的教程。本书由东北大学自然语言处理实验室和小牛翻译联合编写，内容涵盖机器翻译的概论、统计语言建模基础、词法分析和语法分析基础、翻译质量评价等多个方面，为相关领域的高年级本科生和研究生提供了学习资源，同时也为自然语言处理和机器翻译的研究人员提供了参考资料。 ### Key features: - **全面覆盖**：从机器翻译的基础概念到前沿技术，内容系统全面。 - **源代码开放**：所有源代码均已开放，便于学习和研究。 - **教程体系**：分为四个部分，每部分由若干章节组成，章节顺序参考了机器翻译技术发展的时间脉络。 - **参考文献**：包含推荐学习资源，帮助读者深入理解机器翻译。 - **结构清晰**：章节内容按时间顺序和逻辑顺序编排，便于学习和复习。 ### Tech stack: - **编写语言**：TeX - **GitHub**：2794 GitHub stars ### Why it matters: 本书系统地介绍了机器翻译的基础知识和建模方法，对于计算机、人工智能及相关专业领域的高年级本科生和研究生来说，是一本非常有价值的教材。同时，其内容的系统性和开放的源代码也为自然语言处理和机器翻译的研究人员提供了宝贵的参考资料。

Python 2,794 stars

jgravelle/jcodemunch-mcp

Cut AI token costs 95%+ on code exploration. The leading MCP server for precise, symbol-level GitHub code retrieval via tree-sitter AST. Works with Claude Code, Cursor & any MCP client. 313B+ tokens saved.

### Overview: Cut AI token costs by 95%+ on code exploration. jCodeMunch MCP is a leading MCP server that retrieves source code by symbol, not by file, using tree-sitter AST parsing. It works with Claude Code, Cursor, and any MCP-compatible client, saving 313B+ tokens. ### Key features: - Precise symbol-level code retrieval - Tree-sitter AST parsing for byte-level precision - Works with Claude Code, Cursor, VS Code, Codex CLI, and any MCP-compatible client - 96.6% fewer tokens than a grep-and-read agent - 29.2x fewer tokens across 15 lookups ### Tech stack: Python ### Why it matters: Reducing AI token costs by 95%+ can significantly lower operational expenses and help companies avoid AI spend. jCodeMunch MCP is a valuable tool for developers looking to optimize their AI spend and reduce carbon footprint.

Unknown 2,794 stars

merrymercy/awesome-tensor-compilers

A list of awesome compiler projects and papers for tensor computation and deep learning.

### Overview: This repository is a curated list of awesome compiler projects and papers focused on tensor computation and deep learning. It aims to provide developers with a comprehensive overview of various tools and techniques used in optimizing and accelerating machine learning models. ### Key Features: - **Open Source Projects:** A collection of open-source projects that can help in optimizing and accelerating tensor computation. - **Papers:** A selection of papers that provide in-depth insights into compiler design, optimization techniques, and hardware-specific approaches for deep learning. - **Tech Stack:** The repository uses a variety of programming languages and frameworks, including Python, C++, CUDA, and others. - **Why it Matters:** This list is essential for developers looking to improve the performance and efficiency of their deep learning models, especially in terms of hardware acceleration and optimization techniques. ### Tech Stack: Python, C++, CUDA, TensorFlow, PyTorch, Apache TVM, MLIR, XLA, Halide, Glow, nnfusion, Hummingbird, Triton, AITemplate, Hidet, Tiramisu, TensorComprehensions, PlaidML, BladeDISC, TACO, Nebulgym, Speedster, NN-512, DaCeML, Mirage, and many others. ### Why it Matters: This repository serves as a valuable resource for developers looking to enhance the performance of their deep learning models. It provides a curated list of tools and techniques that can help in optimizing and accelerating tensor computation, making it easier to achieve better results on hardware platforms.

Python 2,793 stars

autodistill/autodistill

Images to inference with no labeling (use foundation models to train supervised models).

Overview: Autodistill leverages foundation models to train small, edge-friendly supervised models without human intervention. Key features: - Automatically labels datasets - Trains fast supervised models - Owns your model - Deploy distilled models to the cloud or edge Tech stack: Python Why it matters: Automates the process of training models from unlabeled data to inference, reducing the need for human labeling and improving efficiency. Repository: autodistill/autodistill GitHub stars: 2791

Jupyter Notebook 2,787 stars

Alro10/deep-learning-time-series

List of papers, code and experiments using deep learning for time series forecasting

### Overview: This repository compiles a comprehensive list of state-of-the-art papers, code, and experiments focused on deep learning for time series forecasting. It covers a wide range of methods and techniques, including Autoformer, Long Range Probabilistic Forecasting, Collaborative Graph Neural Networks, and more. The repository also includes classic methods and compares them with deep learning approaches, providing a thorough overview of the field. ### Key Features: - **Autoformer**: A deep learning model for long-term series forecasting. - **Long Range Probabilistic Forecasting**: A method using high-order statistics for probabilistic time series forecasting. - **Online Multi-Agent Forecasting**: A collaborative graph neural network for forecasting. - **End-to-End Learning**: A coherent probabilistic forecasting model for hierarchical time series. - **Neural Basis Expansion Analysis**: Forecasting electricity prices using NBEATSx. - **Autoregressive Denoising Diffusion Models**: A model for multivariate probabilistic time series forecasting. - **CHALLENGES AND APPROACHES**: A survey on challenges and approaches to time-series forecasting in data center telemetry. - **MiniRocket**: A fast deterministic transform for time series classification. - **Learning to Select the Best Forecasting Tasks**: A method for clinical outcome prediction. - **Real-World Anomaly Detection**: Using digital twins and weakly-supervised learning for anomaly detection. - **Inter-Series Attention Model**: A model for forecasting COVID-19 cases. ### Tech Stack: Python, Jupyter Notebook ### Why It Matters: This repository serves as a

Jupyter Notebook 2,787 stars

PhoebusSi/Alpaca-CoT

We unified the interfaces of instruction-tuning data (e.g., CoT data), multiple LLMs and parameter-efficient methods (e.g., lora, p-tuning) together for easy use. We welcome open-source enthusiasts to initiate any meaningful PR on this repo and integrate as many LLM related technologies as possible. 我们打造了方便研究人员上手和使用大模型等微调平台，我们欢迎开源爱好者发起任何有意义的pr！

### Overview: Alpaca-CoT is an instruction-tuning platform designed to unify the interfaces of instruction collection, parameter-efficient methods, and large language models, making it easier for researchers to use and integrate various LLMs. ### Key Features: - Unified Interface for CoT Data, Multiple LLMs, and Parameter-Efficient Methods. - Easy Integration of Various LLMs and Parameter-Efficient Techniques. ### Tech Stack: Jupyter Notebook ### Why It Matters: We have created a platform that simplifies the process of using and integrating large language models, making it accessible to researchers and developers alike.

Python 2,786 stars

Windy3f3f3f3f/claude-code-from-scratch

Build your own Claude Code from scratch. 🔍 Claude Code 开源了 50 万行代码，读不动？用 ~5000 行 TypeScript / Python 从零复现核心架构，11 章分步教程带你理解 coding agent 精髓

Overview: This project is a step-by-step guide to building your own Claude Code from scratch, using approximately 5,000 lines of TypeScript and Python code. It covers the core components such as the Agent Loop, 13 tools, context compression, skill system, multi-agent architecture, and more. The project includes 13 chapters, each with real code that can be run, making it easy to understand the principles of a coding agent. Key features: - Agent Loop - Tools System - System Prompt - CLI & Session - Streaming Output - Permissions - Context Management - Memory System - Skills System - Plan Mode - Multi-Agent Architecture - MCP Integration Tech stack: TypeScript and Python Why it matters: This project is designed to help developers understand the inner workings of a coding agent by building it from scratch. It provides a comprehensive guide with real code examples, allowing readers to easily grasp the principles without needing to read through the 50,000 lines of Claude Code.

Jupyter Notebook 2,783 stars

FurkanGozukara/Stable-Diffusion

FLUX, Stable Diffusion, SDXL, SD3, LoRA, Fine Tuning, DreamBooth, Training, Automatic1111, Forge WebUI, SwarmUI, DeepFake, TTS, Animation, Text To Video, Tutorials, Guides, Lectures, Courses, ComfyUI, Google Colab, RunPod, Kaggle, NoteBooks, ControlNet, TTS, Voice Cloning, AI, AI News, ML, ML News, News, Tech, Tech News, Kohya, Midjourney, RunPod

### Overview: Welcome to the SECourses YouTube channel, where Dr. Furkan Gözükara shares expert-level tutorials on Generative AI. From Stable Diffusion and FLUX to advanced concepts like AGI and Prompt Engineering, this channel is your go-to resource for mastering Generative AI. ### Key Features: - **Stable Diffusion and FLUX**: Comprehensive tutorials on Generative AI models. - **Advanced Topics**: From Prompt Engineering to AGI, this channel covers the latest developments in Generative AI. - **Community Engagement**: Follow SECourses on Discord, Patreon, and Twitter for updates and discussions. ### Tech Stack: - Jupyter Notebook ### Why It Matters: - **Educational Content**: Dr. Gözükara provides in-depth tutorials and guides, ensuring you can master Generative AI from the ground up. - **Community Support**: Engaging with the community through Discord, Patreon, and Twitter helps you stay updated and connected with the latest trends and developments in Generative AI. ### Repository: - **GitHub Repository**: FurkanGozukara/Stable-Diffusion - **Stars**: 2782 ### Primary Language: - **Jupyter Notebook** ### GitHub Description: - **FLUX, Stable Diffusion, SDXL, SD3, LoRA, Fine Tuning, DreamBooth, Training, Automatic1111, Forge WebUI, SwarmUI, DeepFake, TTS, Animation, Text To Video, Tutorials, Guides, Lectures, Courses, ComfyUI, Google Colab, RunPod

Jupyter Notebook 2,783 stars

Visualize-ML/Book6_First-Course-in-Data-Science

Book_6_《数据有道》 | 鸢尾花书：从加减乘除到机器学习；欢迎大家批评指正！纠错多的同学会得到赠书感谢！

Overview: This repository contains a comprehensive guide to data science, focusing on the basics of statistics and machine learning. Key features: - Data Science Fundamentals: Covers the essential concepts of statistics and machine learning. - Practical Examples: Includes real-world applications and examples to illustrate the concepts. - Interactive Jupyter Notebooks: Utilizes interactive Jupyter Notebooks for hands-on learning. - Comprehensive Documentation: Detailed explanations and explanations of the code for better understanding. Tech stack: Jupyter Notebook Why it matters: This guide is essential for anyone looking to understand and apply data science principles, making it a valuable resource for developers and learners alike. Repository: Visualize-ML/Book6_First-Course-in-Data-Science Primary language: Jupyter Notebook GitHub stars: 2783 GitHub description: Book_6_《数据有道》 | 鸢尾花书：从加减乘除到机器学习；欢迎大家批评指正！纠错多的同学会得到赠书感谢！ Readme excerpt: 《统计至简》五折入口：https://zhuanlan.zhihu.com/p/634253719<br> 《数学要素》五折入口：https://zhuanlan.zhihu.com/p/620243026<br> 《矩阵力量》五折入口：https://zhuanlan.zhihu.com/p/634253719<br> 看个人情况，开源资源，永久有效哈。纠错多的同学会得到赠书，

Go 2,783 stars

icereed/paperless-gpt

Use LLMs and LLM Vision (OCR) to handle paperless-ngx - Document Digitalization powered by AI

Overview: paperless-gpt is a tool that integrates Large Language Models (LLMs) with Optical Character Recognition (OCR) to automate the process of generating AI-powered document titles and tags, enhancing document organization and management. Key features: - Harnesses LLMs for better-than-traditional OCR - Links related documents for easy reference and organization - Uses specialized AI OCR services like OpenAI and Google Document AI Tech stack: Go Why it matters: paperless-gpt stands out by supercharging OCR with LLMs, ensuring high accuracy even with tricky scans, and connecting documents that belong together, making document management more efficient and less time-consuming. Repository: icereed/paperless-gpt GitHub stars: 2783

Python 2,783 stars

wzhe06/SparrowRecSys

A Deep Learning Recommender System

Overview: SparrowRecSys is a movie recommendation system based on deep learning, designed to provide a comprehensive learning experience for developers interested in recommendation systems. Key features: - Implements various deep learning models such as Word2vec, DeepWalk, Embedding MLP, Wide&Deep, Neural CF, Two Towers, DeepFM, and DIN. - Utilizes TensorFlow, Spark, and Jetty Server for different components of the recommendation system. - Includes a detailed README with instructions for setting up and running the project, including environment requirements and a quick start guide. Tech stack: Python, TensorFlow 2.0, Spark, Jetty Server. Why it matters: SparrowRecSys offers a practical learning experience for developers interested in deep learning recommendation systems, covering essential components and techniques in a comprehensive manner. Repository: wzhe06/SparrowRecSys Primary language: Python GitHub stars: 2783

Unknown 2,782 stars

RedstoneWill/HsuanTienLin_MachineLearning

Overview: This repository contains all the resources for two of the renowned machine learning courses by Professor Hsuan-Tien Lin from NTU, including video lectures, course notes, books, and detailed materials. Key features: - Video lectures from Professor Hsuan-Tien Lin's courses on coursera. - Comprehensive course notes and detailed materials. - Books: "Learning From Data" by Professor Lin, with additional chapters for the "Machine Learning Techniques" course. - Detailed resources for offline study. - A complete download package for free. Tech stack: Python, Jupyter Notebook, LaTeX, and other programming languages used in the course. Why it matters: These resources provide a comprehensive and detailed overview of machine learning concepts, algorithms, and techniques, making it an essential resource for anyone interested in machine learning.

Go 2,780 stars

postgres-ai/database-lab-engine

DBLab enables 🖖 database branching and ⚡️ thin cloning for any Postgres database and empowers DB testing in CI/CD. This optimizes database-related costs while improving time-to-market and software quality. Follow to stay updated.

Overview: DBLab Engine is a powerful tool designed to optimize database-related costs and improve time-to-market and software quality by enabling database branching and thin cloning for any Postgres database. Key features: - Database branching and thin cloning for any Postgres database - Supports self-managed and managed services like AWS RDS, GCP Cloud SQL, Supabase, and Timescale - Blazing-fast database cloning and branching for efficient development, testing, QA, and staging environments - Covers 100% of database migrations with CI tests Tech stack: Go Why it matters: DBLab Engine optimizes database-related costs while improving time-to-market and software quality by providing fast database cloning and branching capabilities, ensuring 100% coverage of database migrations with CI tests, and supporting various cloud environments and on-premises setups. Repository: postgres-ai/database-lab-engine

Jupyter Notebook 2,775 stars

FasterDecoding/Medusa

Medusa: Simple Framework for Accelerating LLM Generation with Multiple Decoding Heads

Overview: Medusa is a framework designed to accelerate Large Language Model (LLM) generation by utilizing multiple decoding heads, offering a 2.2-3.6x speedup over traditional methods. Key features: - Medusa-1: Optimizes LLM generation for a batch size of 1, achieving a 2x speed increase across various models. - Medusa-2: Supports full-model training, enhancing performance without requiring original training data. - Self-distillation: Integrates Medusa into fine-tuned LLMs for faster generation without needing original training data. Tech stack: Jupyter Notebook Why it matters: Medusa simplifies the acceleration techniques for LLM generation, making them more accessible and efficient, especially for models with limited resources. Repository: FasterDecoding/Medusa GitHub stars: 2775 GitHub description: Medusa: Simple Framework for Accelerating LLM Generation with Multiple Decoding Heads

Python 2,771 stars

tensorflow/ranking

Learning to Rank in TensorFlow

### Overview: TensorFlow Ranking is a comprehensive library designed for Learning-to-Rank (LTR) techniques on the TensorFlow platform, providing essential components for various ranking models. ### Key features: - Commonly used loss functions including pointwise, pairwise, and listwise losses. - Commonly used ranking metrics like Mean Reciprocal Rank (MRR) and Normalized Discounted Cumulative Gain (NDCG). - Multi-item (groupwise) scoring functions. - LambdaLoss implementation for direct ranking metric optimization. - Support for unbiased learning-to-rank from biased feedback data. ### Tech stack: Python ### Why it matters: TensorFlow Ranking offers a convenient and open platform for hosting and advancing state-of-the-art ranking models based on deep learning techniques, facilitating both academic research and industrial applications.

Python 2,769 stars

zilliztech/memsearch

A persistent, unified memory layer for all your AI agents (e.g. Claude Code, Codex, DSH), backed by Markdown and Milvus.

### Overview: MemSearch is a persistent, unified memory layer designed for AI agents like Claude Code, Codex, DSH, OpenClaw, and OpenCode, backed by Markdown and Milvus. ### Key Features: - **Persistent Memory**: MemSearch stores and retrieves information across different AI agents. - **Unified Memory Layer**: Supports multiple AI agents seamlessly. - **Markdown Notes**: Stores and retrieves text-based data using Markdown. - **Milvus Indexing**: Utilizes Milvus for efficient indexing and retrieval. - **Skills from Memory**: Automates the creation of reusable skills from repeated workflows. - **Advanced Memory Maintenance**: Background tasks ensure the memory is up-to-date across sessions. ### Tech Stack: Python ### Why it Matters: MemSearch simplifies the management of AI agent memories, making it easier to maintain and update across different platforms.

Python 2,768 stars

NeptuneHub/AudioMuse-AI

AudioMuse-AI uses sonic analysis to rediscover forgotten songs, uncover hidden connections in your music library, and generate intelligent playlists for Navidrome, Jellyfin, LMS, Lyrion, Emby and Plex: no metadata or external services required.

### Overview: AudioMuse-AI is an opensource and self-hosted tool that uses sonic analysis to rediscover forgotten songs in your music library and generate groove-aware playlists without relying on metadata or external APIs. ### Key features: - **Multiple Music Servers**: Connects several media servers (Navidrome, Jellyfin, LMS, Lyrion, Emby, and Plex) to a single AudioMuse-AI deployment, recognizing the same song across servers and sharing the result. - **Clustering**: Automatically groups sonically similar songs, creating genre-defying playlists based on the music's actual sound. - **Instant Playlists**: Generates a playlist based on your input, like "high-tempo, low-energy music". - **Music Map**: Visualizes your music collection with a vibrant, genre-based 2D map. - **Playlist from Similar Songs**: Finds all the songs in your library that share the same sonic signature as a selected track. - **Song Paths**: Creates a seamless listening journey between two songs. - **Sonic Fingerprint**: Generates playlists based on your listening habits, finding tracks similar to what you've been playing most often. - **Song Alchemy**: Mixes your ideal vibe, adding or subtracting tracks to create a curated playlist. - **Text Search**: Searches your songs with simple text, including mood, instruments, and genre. - **Lyrics Search**: Searches your library by theme, story, or meaning. - **Search by Recording**: Records 20 seconds of whatever is playing around you, or uploads a clip, and

TypeScript 2,768 stars

simonlin1212/Vibe-Research

Vibe-Research：A 股、美股、港股的投研工作台 · 自选与持仓、每日复盘、产业资讯、产业研究、个股深挖、回测、研报库，168 个数据端点。支持 Claude Code、Codex、WorkBuddy 订阅或任意模型 API。 | A research workbench for China A-shares, US and HK stocks.

### Overview: Vibe Research 是一个全面的投研工作台，涵盖了中国A股、美股和港股的分析和研究功能。它提供自选与持仓、每日复盘、产业资讯、产业研究、个股深挖、回测、研报库以及168个数据端点。支持Claude Code、Codex、WorkBuddy订阅或任意模型API。 ### Key features: - **自选与持仓**：实时更新的自选与持仓信息，包括持仓盈亏和当日盈亏。 - **每日复盘**：大盘指数、自选和持仓的详细复盘，涵盖涨跌面、量能、连板高度、两融等指标。 - **产业资讯与研究**：产业资讯和研究，涵盖12个行业的公开信源和近3天的资讯。 - **个股深挖**：个股概览、K线、财报前瞻与复盘、事件回看，以及个股深挖和候选池功能。 - **研报库**：导入PDF、Word等资料，检索、提问，引用能回查原文。 - **回测和持仓体检**：用对话描述想法跑历史回测，结果与指数对比，展示累计收益、回撤、月度、年度4张图，以及按持仓比例回看历史大跌期。 - **支持的AI**：Claude Code、Codex、WorkBuddy订阅，或兼容Chat Completions的模型API。 ### Tech stack: TypeScript ### Why it