Project directory

Projects

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

Unknown 94 stars

HallNautilus24/Anthropics-Portrait-Windows-Setup-Guide

Anthropics Portrait for Windows — step-by-step install and first-run setup on Windows 10 and 11, with download steps in the README.

Overview: This guide provides a step-by-step installation and first-run setup for Anthropics Portrait on Windows 10 and 11, including download instructions. Key features: - Step-by-step installation guide - First-run setup instructions - Download steps provided in README Tech stack: Python, Windows Why it matters: This guide is essential for users looking to set up Anthropics Portrait on their Windows system, ensuring a smooth and straightforward setup process.

Unknown 94 stars

JerryX1110/Awesome-3D-Anomaly-Detection

Awesome-3D/Multimodal-Anomaly-Detection-and-Localization/Segmentation/3D-KD/3D-knowledge-distillation

Overview: This repository contains an extensive dataset for unsupervised anomaly detection and localization in 3D point cloud data. It includes 4147 high-resolution scans from 10 real-world object categories, with a focus on detecting and localizing anomalies in diverse scenarios. Key features: - Comprehensive dataset for anomaly detection and localization in 3D point cloud data. - Contains precise ground truth annotations for anomalies. - Includes supplementary pseudo multimodal features and collaborative discrepancy optimization methods. Tech stack: Python, Open3D, TensorFlow, PyTorch. Why it matters: This dataset and methods provide valuable tools for researchers and developers working on 3D anomaly detection and localization, offering a practical resource for advancing the field.

Python 94 stars

OceanPresentChao/llm-corpus

从零搭建大模型知识库(Build LLM RAG Corpus from scratch)

### Overview: 从零开始构建大模型知识库，实现从中文数据集到向量数据库的全流程。 ### Key features: - 中文数据集处理 - 词向量模型训练 - 文档向量化 - 向量数据库存储知识库 - 本地ChatGLM2-6B大模型部署 - 简单的知识库应用 ### Tech stack: Python ### Why it matters: 本项目提供了一个从零开始构建大模型知识库的完整流程，适用于希望学习和实践大模型知识库搭建的开发者。

Python 94 stars

QuentinWach/image-ranker

Rank images using TrueSkill by comparing them against each other in the browser. 🖼📊

### Overview: Image Ranker is a Python-based project that ranks images using the TrueSkill ELO algorithm. It provides advanced ranking speedups and a web-friendly interface for easy use. The project is maintained and has been starred by 93 developers. ### Key Features: - **TrueSkill ELO Algorithm**: An advanced algorithm that updates global rankings after each comparison, providing accurate results faster. - **Sequential Elimination**: Ranks images in O(N) time by eliminating images with low votes. - **Smart Shuffle**: Minimizes ranking uncertainty with fast shuffles. - **Web GUI**: Easy-to-use interface for browsing and ranking images. - **Export/Save**: Export rankings as CSV and resume from any point. - **Local Directory Support**: Select images from a local directory without needing to upload or move them. ### Tech Stack: Python ### Why it Matters: Image Ranker is useful for applications that require real-time image ranking, such as image galleries, content recommendation systems, or any scenario where image comparisons are needed.

C 94 stars

TDahlmann/canpp

CAN bus analyzing and simulation GUI tool for Windows.

Overview: CAN++ is a free Windows program designed to receive, transmit, and analyze CAN bus messages, including CAN Classic and CAN FD. It supports SuperCAN devices and can be connected via TCP/IP or virtual bus. Key features: - CAN simulation in C++ - CAN traces generation and replay - CAN diagnostics (fault memory, measurement, identification) - CAN statistics - CAN transmit with cyclic, once, and time-offset sending - CAN hardware connection via TCP/IP with SocketCan protocol Tech stack: C Why it matters: CAN++ simplifies the process of working with CAN buses, providing tools for simulation, analysis, and diagnostics, making it essential for developers and engineers working with automotive and industrial control systems. Repository: TDahlmann/canpp GitHub stars: 94 GitHub description: CAN bus analyzing and simulation GUI tool for Windows.

Python 94 stars

Wangmerlyn/MCTS-GSM8k-Demo

This is a repo for showcasing using MCTS with LLMs to solve gsm8k problems

### Overview: This repository showcases the application of Monte Carlo Search Tree (MCTS) with Large Language Models (LLMs) to solve the GSM8k problem. It provides a demo project for a lecture video, demonstrating the implementation and showcasing rough but useful code. ### Key Features: - **Demonstration of MCTS with LLMs for GSM8k Problems** - **Support for `openai` and `deepseek` API calls** - **Token Usage Statistics** - **Command Line Arguments for Customization** - `--provider`: Choose between `openai` or `deepseek` - `--model`: Specify model name (provider-specific if not specified) - `--api-key`: Directly provide API key (reads from environment variables if not specified) - `--base-url`: Specify API base URL (provider-specific if not specified) - `--iterations`: Specify number of MCTS iterations (default 4) ### Tech Stack: - Python ### Why It Matters: This demo project serves as a practical example for students and researchers learning about Monte Carlo Search Tree algorithms and their application with large language models, specifically for solving the GSM8k problem.

Python 94 stars

aadeshkulkarni/sanchay-ai

Takes your video and generates video title, description, hashtags, transcription, subtitles and more.

Overview: Sanchay AI is a generative AI app that converts videos into key elements such as transcriptions, subtitles, and chapters. Key features: - Video transcription - Video subtitles - Video chapter generation Tech stack: Python Why it matters: Sanchay AI offers a valuable service by converting videos into easily accessible content, making it a significant tool for content creators and marketers. Repository: aadeshkulkarni/sanchay-ai Primary language: Python GitHub stars: 94

Jupyter Notebook 94 stars

czczup/UrbanRegionFunctionClassification

第五届百度西安交大大数据竞赛城市区域功能分类 Baseline

Overview: 本项目是第五届百度西安交大大数据竞赛“城市区域功能分类”赛题的参赛方案，使用遥感图像和访问数据两个模态进行特征融合，最终在测试集上获得了0.57的准确率。 Key features: - 使用遥感图像和访问数据两个模态进行特征融合 - 在特征层进行融合，提升了模型的准确率 - 通过数据转换和生成tfrecord，实现了数据的高效处理 Tech stack: Jupyter Notebook Why it matters: 该方案通过结合遥感图像和访问数据，实现了对城市区域功能的准确分类，具有较高的实际应用价值。

Python 94 stars

fyzhang1/FocusGUI

Official code for "Focused Omni-Modal GUI Agents“

Overview: FocusGUI is a lightweight adapter designed for omni-modal GUI agents, focusing on evidence selection and action generation without replacing the pretrained decoder or language model head. Key features: - Channel decomposition - Step-adaptive prioritization - Compact focus - Residual alignment with a frozen decoder Tech stack: Python Why it matters: FocusGUI simplifies the process of integrating multiple modalities into a GUI, ensuring that the most relevant information is prioritized and the user's actions are guided effectively without the need for complex modifications to the original model architecture. Repository: fyzhang1/FocusGUI GitHub stars: 91 GitHub description: Official code for "Focused Omni-Modal GUI Agents"

Swift 94 stars

ghostwright/shadow

Your computer was paying attention the whole time. 14-modality capture. Proactive intelligence. Computer-use training data. Native macOS. All on-device. Open source.

### Overview: Shadow is a personal intelligence engine for macOS that captures and analyzes every signal your computer produces while you work. It generates episodes from your work, runs a continuous heartbeat that pushes proactive observations, and operates vision models and LLMs entirely on Apple Silicon. Shadow is open-sourced and aims to be a comprehensive solution for understanding and improving your computer use. ### Key Features: - **Proactive Intelligence**: Shadow learns from every signal your computer produces and acts on what it learns. - **Native macOS Support**: All features are native to macOS, ensuring compatibility and seamless integration. - **On-Device Processing**: All data is processed locally, ensuring crash-proof recording and minimal impact on system resources. - **Open Source**: Shadow is open-sourced, allowing the community to build on top of what is already here. ### Tech Stack: Swift, Rust, and macOS ### Why It Matters: Shadow is a game-changer for anyone looking to understand and improve their computer use. By capturing every signal your computer produces, Shadow turns raw behavior into structured understanding, and acts on what it learns. Whether you're a developer, a designer, or a general user, Shadow provides valuable insights into how you work, learn, and interact with technology. ### GitHub Stars: 93 ### Repository: ghostwright/shadow

HTML 94 stars

haoheliu/torchsubband

Pytorch implementation of subband decomposition

Overview: This package provides a PyTorch implementation for subband decomposition, transforming waveform into three revertable subband feature representations that can be useful for music source separation and similar tasks. Key features: - Subband decomposition - Reversible subband feature representations - Suitable for music source separation and related tasks Tech stack: PyTorch Why it matters: This package offers a practical solution for transforming audio signals into subband representations, which can enhance the performance of music source separation and related applications.

TypeScript 94 stars

itsmeadarsh2008/opencode-senses

The vision plugin for OpenCode that truly understands images. Inspect, read, and reason about any screenshot or picture with deeper understanding than any other plugin — fully local, private, and free.

### Overview: OpenCode Senses is a plugin for OpenCode that enhances its capabilities by understanding and analyzing images. This plugin provides a deeper level of insight into screenshots and images, making it easier for text-only coding models to reason about and interpret visual data. ### Key Features: - **Image Analysis**: Identifies objects, colors, and locations within images. - **Structured Evidence**: Returns structured data that can be easily reasoned over by text models. - **No API Keys Needed**: No need for external APIs or hidden costs. - **Privacy**: The plugin is designed to be fully local and private, ensuring no pictures of your desktop leave the machine. ### Tech Stack: TypeScript ### Why It Matters: OpenCode Senses transforms OpenCode into a powerful tool for text-only coding models, making it easier to understand and reason about visual data. This plugin enhances the capabilities of OpenCode, making it more versatile and useful for various coding tasks.

Jupyter Notebook 94 stars

jhj0517/finetuning-notebooks

### Overview: This repository contains a collection of Jupyter Notebook-based fine-tuning notebooks designed for training LoRA models. These notebooks are optimized for use with Google Colab and can be applied to various models, including Flux, Hunyuan Video, Wan Video, SDXL, LTX Video, and a Diffusers Converter (Util) utility. ### Key Features: - **LoRA Models**: Fine-tuning LoRA models for different applications, such as DreamBooth, SDXL, LTX Video, and a utility for Diffusers Converter. - **Dataset Sizing**: Typically, 5 to 20 images are sufficient for training LoRA models. - **Colab Integration**: All notebooks are compatible with Google Colab, allowing for seamless training and experimentation. ### Tech Stack: The repository uses Jupyter Notebook for its development and execution. ### Why It Matters: These notebooks provide a practical and efficient way to fine-tune LoRA models, making them accessible for researchers and developers working on image generation and text-to-image conversion tasks. The compatibility with Colab ensures that users can easily run and experiment with the models without needing to install additional software.

Python 94 stars

kyegomez/PALI

Democratization of "PaLI: A Jointly-Scaled Multilingual Language-Image Model"

Overview: PALI is an open-source implementation of a Multi-Modality AI model designed to scale language-image learning across 100+ languages. This project aims to democratize access to advanced AI technologies by providing a robust and efficient model. Key features: - Democratized access to advanced AI technologies - Multi-Modality AI model - Supports 100+ languages - Compatible with Tokenizer and VIT model Tech stack: Python Why it matters: PALI offers a significant contribution to the democratization of AI by providing a versatile and scalable model that can be easily integrated into various applications, making it accessible to a wider audience. Its ability to handle multilingual data and support a wide range of languages makes it a valuable tool in the field of natural language processing and computer vision. Repository: kyegomez/PALI Primary language: Python GitHub stars: 94 GitHub description: Democratization of "PaLI: A Jointly-Scaled Multilingual Language-Image Model" GitHub issues: [![GitHub issues](https://img.shields.io/github/issues/kyegomez/pali)](https://github.com/kyegomez/pali/issues) GitHub forks: [![GitHub forks](https://img.shields.io/github/forks/kyegomez/pali)](https://github.com/kyegomez/pali/network) GitHub stars: [![GitHub stars](https://img.shields.io/github/stars/kyegomez/pali)](https://github.com/kyegomez/pali/stargazers) GitHub

Unknown 94 stars

mani-kantap/llm-inference-solutions

A collection of all available inference solutions for the LLMs

### Overview: This repository contains a collection of inference solutions for large language models (LLMs). It serves as a comprehensive resource for developers looking to optimize and deploy LLMs efficiently. ### Key Features: - **vLLM**: High-throughput and memory-efficient inference and serving engine for LLMs. - **Text-Generation-Inference**: Efficient and scalable text generation inference for LLMs. - **llm-engine**: Scale LLM Engine public repository for efficient inference. - **DeepSpeed**: Deep learning optimization library for easy, efficient, and effective distributed training and inference. - **OpenLLM**: Operating LLMs in production with ease. - **LMDeploy**: Toolkit for compressing, deploying, and serving LLMs. - **FlexFlow**: Distributed deep learning framework with automatic parallelization. - **CTranslate2**: Fast inference engine for Transformer models. - **FastChat**: Open platform for training, serving, and evaluating large language models. - **Triton Inference Server**: Optimized cloud and edge inferencing solution. - **Lepton.AI**: Pythonic framework to simplify AI service building. - **ScaleLLM**: High-performance inference system for LLMs designed for production environments. - **Lorax**: Serve hundreds of fine-tuned LLMs in production for the cost of one. - **TensorRT-LLM**: Provides users with an easy-to-use Python API to define LLMs and build TensorRT engines. - **mistral.rs**: Blazingly fast LLM inference. - **NanoFlow

Jupyter Notebook 94 stars

markweberdev/maskbit

Implementation of the paper "MaskBit: Embedding-free Image Generation from Bit Tokens"

### Overview: MaskBit is an implementation of the paper "MaskBit: Embedding-free Image Generation via Bit Tokens," which presents a modernized VQGAN+ model and a novel embedding-free image generation framework. This approach achieves state-of-the-art performance on the ImageNet 256×256 class-conditional image generation benchmark. ### Key features: - **VQGAN+ Model**: A shared representation model that uses a shared representation in the tokenizer and generator. - **Bit Tokens**: Leveraging bit tokens for embedding-free image generation. - **Improved Tokenizer**: A publicly available, reproducible, and high-performing VQGAN model. - **State-of-the-Art Performance**: Achieves a significant improvement over the original VQGAN developed three years ago. - **Modern Lookup-Free Quantization (LFQ)**: Analyzed the latent representation and observed highly structured semantics for bit token representation. ### Tech stack: - **Jupyter Notebook**: Used for implementation and reproducibility. ### Why it matters: MaskBit provides a modernized and improved approach to image generation, offering significant improvements in performance while using a significantly smaller model compared to autoregressive models.

Python 94 stars

parlance-zz/dualdiffusion

Dual Diffusion is a generative diffusion model for music trained on video game soundtracks.

Overview: Dual Diffusion is a generative diffusion model designed for music, trained on video game soundtracks from the mid-90s to the present day. Key features: - Uses a custom architecture based on the EDM2 UNet and a custom 2-stage VAE design. - Trains on CLAP audio embeddings and is conditioned on raw audio. - Supports both 1D and 2D formats, with 2D formats being more compute and parameter efficient. - Utilizes separable attention for high-dimensional models, improving performance without sacrificing quality. - Incorporates point-wise loss, multi-scale spectral loss, and phase loss to enhance model performance. - Uses uniform variance in latent space and lower target SNR for improved latent diffusion model performance. Tech stack: Python Why it matters: Dual Diffusion offers a unique approach to generative diffusion models, focusing on music generation from video game soundtracks. The model's architecture and training techniques are designed to produce high-quality, musically coherent music, making it a valuable tool for artists and developers in the audio and music production fields.

Python 94 stars

robbiewatt1/ClimateDiffuse

Diffusion for climate downscaling

### Overview: This repository contains code for a diffusion-based model applied to climate downscaling, developed by Robbie A. Watt and Laura A. Mansfield. The model is based on a diffusion process and a UNet, and it has been adapted from the implementation by T. Karras et al. and the original code from NVlabs/edm. ### Key features: - Diffusion-based model for climate downscaling - UNet architecture for enhanced performance - Adapted from T. Karras et al. and NVlabs/edm - Code for training, inference, and plotting ### Tech stack: Python, PyTorch, Tensorboard, Xarray, NetCDF4, Cartopy, Matplotlib, SciPy, NumPy ### Why it matters: This project showcases the application of diffusion models and UNets in climate downscaling, providing a novel approach to downscaling climate data, which is crucial for improving climate predictions and understanding climate change impacts.

Python 94 stars

rudyoactiv/typescribe-handwriting

⚡ Create handwritten documents from text with a Neural Network!

Overview: TypeScribe is a tool that converts typed text into realistic handwriting using Machine Learning, making it easy to create organic documents like Christmas cards and essays. Key features: - Realistic Handwriting Generation with a Recurrent Neural Network (RNN) - Choose from 12 predefined handwriting styles - Adjust line spacing, page size, margins, ink color, pen thickness, and more - User-friendly interface - Automatically splits large texts into multiple lines, paragraphs, and pages - Scalable SVG Output that maintains quality even on resizing Tech stack: Python Why it matters: TypeScribe simplifies the process of creating personalized documents, making it accessible for anyone looking to send out organic Christmas cards or complete essays with a touch of creativity.

Python 94 stars

shawn0728/Unify-Agent

🐧 Unify-Agent: An end-to-end unified multimodal agent for faithful, knowledge-grounded image generation.

### Overview: Unify-Agent is an end-to-end unified multimodal agent designed for faithful, knowledge-grounded image generation. Unlike conventional text-to-image models, it can actively access external world knowledge at inference time, enabling more accurate and faithful image synthesis. ### Key Features: - **Unified Multimodal Agent**: Combines understanding and generation capabilities in a single model. - **THINK**: Identifies missing knowledge from prompts. - **RESEARCH**: Retrieves relevant textual and visual evidence. - **RECAPTION**: Converts retrieved knowledge into structured guidance. - **GENERATE**: Synthesizes final grounded images. ### Tech Stack: Python ### Why It Matters: Unify-Agent significantly improves factual visual synthesis, outperforming strong open-source baselines across benchmarks like FactIP, WiSE, KiTTEN, and T2I-FactualBench. This work highlights a new paradigm for text-to-image generation, moving from closed-book generation to open-book, agentic generation where models actively reason over external knowledge before synthesis.