Project directory

Projects

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

Java 300 stars

houbb/nlp-hanzi-similar

The hanzi similar tool.(汉字相似度计算工具，中文形近字算法。可用于手写汉字识别纠正，文本混淆等。)

Overview: This project provides a tool for calculating the similarity between Chinese characters, focusing on the algorithm for comparing similar characters within the Hanzi set. It aims to contribute to the field of Chinese Natural Language Processing by offering a basic similarity calculation tool. Key features: - Supports calculating the similarity between two Chinese characters. - Allows for customization of the algorithm to fit various applications. - Provides a way to define custom weights for the algorithm. - Enables returning a list of similar characters for a single character. - Offers a method to return a list of similar characters for multiple characters. Tech stack: Java Why it matters: This project serves as an example of how to implement a basic similarity calculation tool for Chinese characters, which can be useful for applications such as hand-written character recognition and text confusion correction.

Python 300 stars

lucidrains/routing-transformer

Fully featured implementation of Routing Transformer

### Overview: Routing Transformer is a fully featured implementation of the Routing Transformer model proposed in the paper "Routing Transformer" by Wang et al. The model uses k-means clustering to route similar queries and keys into the same cluster for attention, improving efficiency and performance. It is designed to be fully integrated into existing language models, making it easy to use and deploy. ### Key features: - **Routing Clustering**: Uses k-means clustering to route similar queries and keys into the same cluster for attention. - **Efficient Attention**: Reduces the number of attention heads by clustering similar queries and keys, improving efficiency. - **Flexible Parameters**: Offers various parameters for tuning, including embedding dimension, attention dropout, and more. - **Causal Model**: Supports causal language models, making it suitable for tasks like language modeling and text generation. ### Tech stack: Python ### Why it matters: Routing Transformer is a significant improvement over traditional transformer models, particularly in terms of efficiency and performance. By clustering similar queries and keys, it reduces the number of attention heads, leading to faster computation and better resource utilization. This makes it an excellent choice for applications where computational efficiency is crucial, such as large language models and text generation tasks.

Python 300 stars

lyonzin/knowledge-rag

Local RAG MCP server for Claude Code — hybrid search (semantic + BM25), cross-encoder reranking, 13 MCP tools, 20 format parsers. Zero external servers, zero API keys.

### Overview: knowledge-rag is a local RAG server designed for Claude Code, Cursor, and other AI agents. It offers hybrid search, cross-encoder reranking, 35 file formats support, and zero external servers and API keys, ensuring a zero-downtime reindex and enterprise-grade plumbing. ### Key features: - Hybrid search - Cross-encoder reranking - 13 MCP tools - 20 format parsers ### Tech stack: Python ### Why it matters: knowledge-rag is a zero-cloud, zero-api-key solution for local RAG servers, making it ideal for developers and enterprises looking to integrate AI agents with Claude Code and other MCP clients.

Python 300 stars

radi-cho/datasetGPT

A command-line interface to generate textual and conversational datasets with LLMs.

Overview: datasetGPT is a command-line interface and Python library designed for generating textual and conversational datasets using Large Language Models (LLMs). It supports inference across multiple LLMs, including OpenAI, Cohere, and Petals, and can be used for various applications such as training detectors, collecting datasets for research, automating tasks, and fine-tuning models. Key features: - Supports multiple LLMs for generating textual datasets - Allows for the customization of prompts and options for different LLMs - Provides a CLI interface for easy command execution - Outputs in a JSON format for easy parsing and analysis Tech stack: Python Why it matters: datasetGPT simplifies the process of generating datasets for research and application development, making it a valuable tool for researchers, developers, and anyone looking to leverage LLMs for text generation and analysis. Repository: radi-cho/datasetGPT Primary language: Python GitHub stars: 300 GitHub description: A command-line interface and Python library for inferencing Large Language Models to generate textual and conversational datasets.

Unknown 299 stars

Thinklab-SJTU/Awesome-LLM4EDA

### Overview: This repository is dedicated to resources and projects utilizing Large Language Models (LLMs) for Electronic Design Automation (EDA). It includes a list of papers, tools, and datasets that can help developers and researchers in the field. ### Key Features: - **Assistant Chatbot:** A chatbot powered by LLMs for knowledge acquisition and Q&A, enhancing user interaction with EDA software. - **HDL and Script Generation:** LLMs generate RTL codes and EDA scripts for complex design tasks. - **Code Analysis and Verification:** LLMs are used for bug detection, code summarization, and security checks. ### Tech Stack: Unknown ### Why it Matters: This repository provides a comprehensive list of resources and projects that leverage LLMs for EDA, offering valuable insights and tools for developers and researchers in the field of EDA.

Python 299 stars

asigalov61/Tegridy-MIDI-Dataset

Tegridy MIDI Dataset for precise and effective Music AI models creation.

### Overview: The Tegridy MIDI Dataset is a comprehensive, multi-instrumental symbolic music resource designed for Music Information Retrieval (MIR) and the development of Music AI models. It includes curated MIDI collections such as *Tegridy MIDI Dataset*, *Tegridy Piano*, *Tegridy Piano-Violin*, and *Tegridy Children Songs*, all released under the CC BY-NC-SA license. The repository functions as a central hub, cataloging and linking to external resources like massive corpora, specialized datasets, and software tools. ### Key features: - **Original, Normalized MIDI Collections**: High-quality MIDI files from various instruments, including piano, violin, and children's songs. - **Massive Datasets**: Over 6.74 million MIDI files from Discover, Godzilla, and Meta, designed for large-scale MIR and symbolic music AI. - **Specialized Research Datasets**: Small, high-quality datasets for specific research tasks, such as MAESTRO, GiantMIDI, and POP909. - **Thematic Collections**: Smaller, community-driven collections focusing on specific genres, games, or artists, like Touhou MIDI Collection, wild-west-midi-2, and classical-music-midi. ### Tech stack: Python ### Why it matters: By providing a unified platform combining original data, external references, and practical tooling, the Tegridy MIDI Dataset aims to lower the barrier to entry and accelerate research in symbolic music processing, multi-instrumental music generation,

Python 299 stars

lucidrains/nGPT-pytorch

Quick implementation of nGPT, learning entirely on the hypersphere, from NvidiaAI

### Overview: nGPT-pytorch is a quick implementation of nGPT, a normalized Transformer model designed to learn on the hypersphere. This model is based on NvidiaAI's research, aiming to explore the potential expressivity loss during training. ### Key features: - **nGPT Model**: A normalized Transformer architecture that learns on the hypersphere. - **Continual Learning**: The model is designed to study its performance in the context of continual learning. - **Continual Learning with Vision Transformers**: Adaptation of the model to vision transformers. - **Off-Policy Reinforcement Learning**: Improved UTD (Unsupervised Transition Dynamics) using normalized feedforward. ### Tech stack: Pytorch ### Why it matters: The model's ability to learn on the hypersphere and its potential applications in continual learning and off-policy reinforcement learning make it a valuable addition to the developer community.

Python 299 stars

opendatalab/MinerU-HTML

MinerU-HTML: An SLM-powered HTML main content extractor that outputs clean HTML bodies. Perfect for Deep Research Agents, RAG applications, and training data generation.

Overview: MinerU-HTML is an advanced HTML main content extraction tool powered by Small Language Models (SLM). It efficiently identifies and extracts clean HTML bodies from complex web pages, perfect for Deep Research Agents, RAG applications, and training data generation. Key features: - Accurate main content extraction - Extensible output with [MinerU-Webkit](https://github.com/opendatalab/MinerU-Webkit) for conversion to Markdown, JSON, and TXT - Compact format for faster inference - Regex structured output - Comprehensive processing pipeline - Multiple inference backend support - Fault tolerance and modular design - Complete testing Tech stack: Python Why it matters: MinerU-HTML is a powerful tool for extracting clean HTML bodies from complex web pages, making it ideal for Deep Research Agents, RAG applications, and training data generation.

Python 299 stars

shawn0728/OpenSearch-VL

🔍 OpenSearch-VL provides a fully open recipe for training strong multimodal deep search agents through high-quality data curation, diverse visual/search tools, and fatal-aware agentic reinforcement learning.

### Overview: OpenSearch-VL provides a comprehensive framework for training multimodal deep search agents using agentic reinforcement learning. It includes open-source data, code, and model checkpoints to enable reproducible research. ### Key Features: - **Data:** Open-source curation pipeline for image-grounded multi-hop VQA. - **Tools:** Unified visual and retrieval tool environment for seamless integration. - **Algorithm:** Multi-turn fatal-aware GRPO for handling cascading failures. ### Tech Stack: Python ### Why it Matters: This release closes the gap in training frontier multimodal search agents, making it easier for researchers to reproduce top-tier models and accelerate innovation in the field. ### Repository: shawn0728/OpenSearch-VL ### GitHub Stars: 296 ### Primary Language: Python ### GitHub Description: 🔍 OpenSearch-VL provides a fully open recipe for training strong multimodal deep search agents through high-quality data curation, diverse visual/search tools, and fatal-aware agentic reinforcement learning.

Python 299 stars

wangchuxiaoji-oss/doubao2api

Reverse-engineered Doubao (豆包) API → OpenAI-compatible REST service. Free multimodal chat, image/video/music generation, and file hosting for AI agents.

### Overview: 本项目是一个逆向工程的豆包（Doubao）API客户端，旨在为任何纯文本模型提供免费的多模态能力。通过 OpenAI 兼容的接口，用户可以访问包括识图、读文件、生成图片、音乐、视频等在内的多种功能。 ### Key features: - 多模态对话 - 识图功能 - 读取多种文件格式 - 生成图片、音乐、视频 - 文件上传与下载服务 ### Tech stack: Python ### Why it matters: 这个项目为任何需要多模态能力的纯文本模型提供了一种免费的解决方案，使得它们能够像OpenAI的模型一样进行多模态交互，从而极大地扩展了它们的使用场景和功能。

TypeScript 299 stars

yonatangross/orchestkit

The Complete AI Development Toolkit for Claude Code. 106 skills, 36 agents, 171 hooks. Install `ork` for stable (v9.x), or `ork-alpha` for the v10 line, which ships daily.

### Overview: OrchestKit is a comprehensive plugin for Claude Code that includes skills, agents, and hooks, designed to streamline development and testing processes. ### Key Features: - **Skills**: 115 skills across various domains. - **Agents**: 36 agents for automating tasks. - **Hooks**: 161 hooks for integrating with existing systems. ### Tech Stack: TypeScript ### Why It Matters: OrchestKit simplifies the development process by providing a unified interface for skills, agents, and hooks, making it easier to manage and integrate with Claude Code. ### GitHub Stars: 298

Python 298 stars

AMontgomerie/question_generator

An NLP system for generating reading comprehension questions

Overview: Question Generator is an NLP system designed to generate reading comprehension-style questions from various text sources such as news articles or book excerpts. It utilizes pretrained models from HuggingFace Transformers to create high-quality questions. Key features: - Generates mixed style questions (full-sentence and multiple choice) - Supports three answer styles: sentences, multiple choice, and a mix of both - Outputs question-answer pairs with full sentences, named entities, or both - Can be instantiated and used programmatically for easy integration Tech stack: Python Why it matters: This system is valuable for educational applications, improving reading comprehension skills, and enhancing the quality of questions generated by NLP models.

R 298 stars

JamesHWade/gpttools

gpttools extends gptstudio for package development to help you document code, write tests, or even explain code

### Overview: gpttools extends gptstudio for R package development, helping with documentation, testing, and code explanation. ### Key Features: - Enhanced package documentation with LLMs - Improved test automation using LLMs - Code explanation capabilities for better readability ### Tech Stack: R ### Why it Matters: gpttools simplifies integrating large language models into R package development workflows, making it easier to incorporate AI into project processes.

TypeScript 298 stars

Tenney95/AI-Canvas-tauri

Local-first AI canvas for visual workflows, AI short drama, image/video generation, storyboarding and asset management. ComfyUI, AI agents, MCP & Blender. 本地优先 AI 画布：AI短剧、AI资产管理、图像/视频生成、分镜制作与视频剪辑。

### Overview: AI Canvas Tauri 是一个本地优先的AI画布，集成了ComfyUI、AI对话助手和MCP，提供从创意到视频的完整创作流程，包括AI短剧创作、AI资产管理、图像/视频生成和分镜制作。它支持多种模态节点和多语言界面，适用于个人和团队使用。 ### Key features: - **AI短剧创作**：无缝连接ComfyUI、AI对话助手和MCP，实现从创意到视频的完整创作流程。 - **AI资产管理**：统一管理角色库、人物、场景、道具和本地素材，支持全局资产和自动同步。 - **图像/视频生成**：生成高质量的图像、视频和动画，支持多种模型和格式。 - **分镜制作**：支持编写和修改分镜，准备配音和视频素材，推送剪辑时间轴。 - **本地存储与同步**：在本地保存项目和媒体，支持凭据存储和自动同步。 - **角色与创作资产**：管理项目与全局人物、场景、道具和武器，支持多选和跨项目复制。 - **MCP 外部控制**：允许外部AI工具通过本地stdio或Streamable HTTP控制画布、媒体和剪辑项目。 - **3D导演台**：提供轻量级导演台和Blender高级编辑功能，用于构建场景和预演镜头。 - **多语言界面**：支持简体中文、English、日本語和한국어，

Python 298 stars

TransformerOptimus/AutoNode

AutoNode: A Neuro-Graphic Self-Learnable Engine for Cognitive GUI Automation

Overview: AutoNode is a self-operating computer system designed to automate web interactions and data extraction processes. It uses advanced technologies like OCR, YOLO for object detection, and a custom site-graph to navigate and interact with web pages programmatically. Key features: - OCR (Optical Character Recognition) - YOLO (You Only Look Once) models for object detection - Custom site-graph for web page navigation Tech stack: Python Why it matters: AutoNode is a powerful tool for automating cognitive GUI interactions, making it easier and more efficient to extract data and automate tasks on websites. Its ability to navigate complex web pages and perform intricate interactions makes it a valuable asset in various development and automation projects. Repository: TransformerOptimus/AutoNode Primary language: Python GitHub stars: 298

Python 298 stars

alphanome-ai/sec-parser

Parse SEC EDGAR HTML documents into a tree of elements that correspond to the visual (semantic) structure of the document.

### Overview: This project parses SEC EDGAR HTML documents into a tree of elements that correspond to the visual (semantic) structure of the document. It is designed to help developers understand the structure of the SEC filings by visualizing them in a tree format. ### Key Features: - Parses SEC EDGAR HTML documents. - Converts the documents into a tree structure. - Corresponds the visual structure of the document to the elements in the tree. - Supports various Python libraries for parsing and rendering. ### Tech Stack: - Python ### Why It Matters: This project is crucial for developers who need to quickly understand the structure of SEC filings. By visualizing the documents in a tree format, it simplifies the process of analyzing and extracting information from the filings. ### Repository: - **Repository:** alphanome-ai/sec-parser - **Stars:** 296 - **Description:** Parse SEC EDGAR HTML documents into a tree of elements that correspond to the visual (semantic) structure of the document. ### GitHub Actions: - **CI/CD:** GitHub Actions workflows for continuous integration and continuous deployment. - **Health:** GitHub Actions workflows for monitoring the health of the project. - **Stability:** GitHub Actions workflows for determining the stability level of the project. ### Quality: - **Type Checking:** Pyright (type-checked). - **Code Style:** Black (code style enforced). - **Coverage:** Ruff (code coverage). - **Downloads:** PyPI downloads. ### Distribution: - **Version:** PyPI version. - **Python Version:** PyPI Python

TypeScript 298 stars

helixnow/deep-student

An open-source, local-first AI learning workbench

### Overview: DeepStudent is an open-source, local-first AI learning workbench designed to enhance the study process. It integrates seamlessly with various apps on your Study Desktop, automating tasks and providing personalized learning experiences. ### Key Features: - **Study Agent**: Works directly in every app on your Study Desktop, generating notes, mind maps, exam sets, flashcards, and more. - **Long-Term Tasks**: Handles research, organizing, and question writing without manual intervention, ensuring tasks are completed efficiently. - **Controlled Actions**: Approved by risk level, allowing for easy undoing of notes and edits to the Study Desktop. - **Personalized Learning**: Remembers weak spots and study habits, with built-in skills and model providers for a variety of features. ### Tech Stack: TypeScript ### Why It Matters: DeepStudent simplifies the learning process by automating repetitive tasks, providing personalized learning experiences, and ensuring tasks are completed efficiently. Its integration with various apps makes it a versatile tool for any student.

Unknown 298 stars

kahne/NonAutoregGenProgress

Tracking the progress in non-autoregressive generation (translation, transcription, etc.)

### Overview: Non-Autoregressive Generation Progress tracks the evolution of non-autoregressive generation techniques, focusing on translation, transcription, and other applications. This project aims to explore diverse methods for improving the efficiency and accuracy of non-autoregressive models. ### Key Features: - **Non-Autoregressive Translation**: Techniques for generating translations without relying on autoregressive models. - **Non-Autoregressive Machine Translation**: Methods for improving the quality of machine translation by non-autoregressive approaches. - **Non-Autoregressive Text Generation**: Approaches for generating text without the need for autoregressive models. - **Non-Autoregressive Neural Machine Translation**: Techniques for enhancing the performance of neural machine translation models. ### Tech Stack: The project utilizes a combination of Python, TensorFlow, and other machine learning libraries for implementing and evaluating non-autoregressive generation models. ### Why It Matters: Non-autoregressive generation methods offer significant improvements in terms of speed and efficiency, making them particularly useful in applications where real-time processing is crucial. These techniques help in reducing computational complexity and improving the overall performance of models, thereby enhancing their applicability in various domains such as text generation, translation, and transcription.

Unknown 298 stars

shengchaochen82/Awesome-Foundation-Models-for-Weather-and-Climate

A comprehesive survey about foundation models for weather and cliamte data understanding.

### Overview: This repository is a comprehensive survey on foundation models for weather and climate, focusing on time series and textual data. It aims to summarize the recent advances in this field, providing up-to-date resources, and highlighting key areas and challenges. ### Key Features: - **Time Series Data Analysis**: Techniques and models for analyzing time series data. - **Textual Information Processing**: Methods for understanding and interpreting textual data related to weather and climate. - **Model Architectures**: Overview of various architectures used in foundation models for weather and climate. - **Application Scopes**: Different applications of these models in weather and climate data understanding. - **Task-Specific Datasets**: Examples of datasets used for specific tasks in this field. ### Tech Stack: Python, Jupyter Notebooks, and some basic data manipulation tools. ### Why It Matters: This survey provides a valuable resource for researchers and practitioners looking to enhance the accuracy of weather and climate data analysis through the use of foundation models. It offers insights into the latest advancements and challenges in the field, guiding future research directions. ### Repository: shengchaochen82/Awesome-Foundation-Models-for-Weather-and-Climate - GitHub Stars: 299 - GitHub Description: A comprehesive survey about foundation models for weather and climate data understanding. - Primary Language: Unknown - GitHub URL: [shengchaochen82/Awesome-Foundation-Models-for-Weather-and-Climate](https://github.com/shengchaochen82/Awesome-Foundation-Models-for-Weather-and

Unknown 297 stars

30lm32/ml-projects

ML based projects such as Spam Classification, Time Series Analysis, Text Classification using Random Forest, Deep Learning, Bayesian, Xgboost in Python

### Overview: This repository contains a collection of machine learning and data engineering projects that I have been working on. The projects range from classification tasks like spam SMS detection and novel classification to time series analysis and deep learning models. Each project is accompanied by detailed documentation and code snippets, making it an excellent resource for developers looking to learn and implement various ML techniques. ### Key Features: - **Spam SMS Classification**: Uses Naive Bayes, SVM, Random Forest Classifier, and Deep Learning models. - **Fashion MNIST with Tensorboard and MLflow**: Integrates a CNN model into Tensorboard and MLflow for better visualization and tracking. - **Multi-language and Multi-label Classification**: Utilizes LightGBM and TF-IDF for classification tasks. - **Time Series Analysis**: Applies SARIMAX models to analyze time series data. - **Prediction Skip Action**: Uses LightGBM, Linear Regression, and Logistic Regression for prediction tasks. - **Hairstyle Classification**: Applies LightGBM and TF-IDF for classification. - **Dockerized Apache Flink Application**: Integrates an Apache Flink application with Docker for better deployment and scalability. - **Crawler as a Service**: Implements DFS and BFS for crawling services. - **Forecasting Promos Impact**: Uses Random Forest Regressor and ARIMA/SARIMAX models for forecasting. - **Random Forest Classification in PySpark**: Provides a tutorial on implementing Random Forest Classifier in PySpark. ### Tech Stack: - **Python**: The primary programming language used for implementing ML models and data analysis. - **TensorFlow**: