AI Trending

Project directory

Projects

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

Python 1,727 stars

google-research/torchsde

Differentiable SDE solvers with GPU support and efficient sensitivity analysis.

### Overview: This library provides efficient, GPU-accelerated solvers for differentiable stochastic differential equations (SDEs) with Python bindings. It supports both Ito and Stratonovich formulations and includes support for efficient sensitivity analysis, making it suitable for various applications in machine learning and physics simulations. ### Key features: - **GPU Support**: Solves SDEs on the GPU for faster computation. - **Efficient Sensitivity Analysis**: Computes sensitivities of the solution with respect to the parameters efficiently. - **Differentiable SDEs**: Provides solvers that can be used in backpropagation for training neural networks. ### Tech stack: Python ### Why it matters: This library simplifies the process of solving and training SDEs, making it easier to integrate these models into various applications such as reinforcement learning, physics simulations, and more. The GPU support and efficient sensitivity analysis make it particularly useful for real-time applications and complex simulations where speed and accuracy are critical.
Python 1,726 stars

Lightning-Universe/lightning-flash

Your PyTorch AI Factory - Flash enables you to easily configure and run complex AI recipes for over 15 tasks across 7 data domains

Overview: Lightning-Universe/lightning-flash is a comprehensive PyTorch AI Factory designed to simplify the configuration and execution of complex AI recipes across 15 tasks in 7 data domains. Key features: - Supports over 15 tasks across 7 data domains - Offers pre-trained backbones and heads for various tasks - Easy installation via PyPI - Documentation and community support available Tech stack: Python Why it matters: This framework provides a production-grade solution for researchers and developers, making it accessible and efficient for AI experimentation and deployment.
TypeScript 1,726 stars

legeling/PromptHub

一款包含了 Prompt管理，Skill管理，Agent管理的一站式AI工具箱，助你高效管理提示词，一键分发skills ，一站式管理Agent资产，并实现云同步，备份，版本管理 | An all-in-one AI toolbox for prompt, agent, and skills management. Reuse prompts, distribute skills with one click, manage agent assets, and support cloud sync, backup, and version control

Overview: PromptHub 是一个集 Prompt、Skill 和 Agent 管理于一体的 AI 工具箱，旨在帮助开发者高效管理提示词、一键分发技能并实现云同步、备份和版本控制。 Key features: - 提供 Prompt 管理功能 - 实现技能一键分发 - 支持 Agent 资产管理 - 支持云同步、备份和版本控制 Tech stack: TypeScript, Electron, React, TailwindCSS, SQLite Why it matters: 通过 PromptHub，开发者可以轻松管理提示词、技能和 Agent 资产，实现多模型测试和版本控制，提高工作效率并确保数据安全。
Python 1,723 stars

Lynpoint/CyberVerse

Self hosted, real-time digital human agent platform. Build voice-first AI agents with WebRTC, persona memory, tools, RAG, and optional digital-human video.

### Overview: CyberVerse is a self-hosted, real-time digital-human agent platform that enables developers to build voice-first AI agents using WebRTC, persona memory, tools, RAG, and optional digital-human video capabilities. ### Key Features: - Real-time digital-human video interaction - Persona memory integration - Tools for customization and development - Rich media capabilities including RAG and optional digital-human video - WebRTC for seamless, real-time communication ### Tech Stack: Python ### Why it Matters: CyberVerse provides a powerful framework for developers to create interactive digital-human agents, enhancing user experience through natural voice interaction and real-time video capabilities. Repository: Lynpoint/CyberVerse GitHub Stars: 1723 GitHub Description: Self-hosted, real-time digital-human agent platform. Build voice-first AI agents with WebRTC, persona memory, tools, RAG, and optional digital-human video capabilities.
Python 1,723 stars

meta-pytorch/tnt

A lightweight library for PyTorch training tools and utilities

Overview: TNT is a lightweight library for PyTorch that provides training tools and utilities. Key features: - Training tools and utilities - Easy integration with PyTorch - Cross-platform compatibility Tech stack: Python Why it matters: TNT simplifies the training process by offering a set of tools and utilities that can be easily integrated into PyTorch projects, making it a valuable resource for developers working with deep learning models.
PHP 1,722 stars

LLPhant/LLPhant

LLPhant - A comprehensive PHP Generative AI Framework using OpenAI GPT 4. Inspired by Langchain

### Overview: LLPhant is a comprehensive PHP library designed for generating artificial intelligence (Gen AI) and vector database integration. It leverages OpenAI's GPT-4 and supports various LLMs, making it compatible with Symfony and Laravel. The library aims to simplify the development of powerful applications by providing a straightforward interface. ### Key Features: - **JevClassifier with Typed Classification**: Supports Jev classification with Noul, Choice, and Score rubrics, including malicious prompt detection. - **LLM Support**: Compatible with OpenAI, Anthropic, Mistral, Ollama, LLMman, LM Studio, and services compatible with the OpenAI API. - **Vector Databases Integration**: Seamlessly integrates with vector databases for enhanced search capabilities. - **Composer Installation**: Easy to install via Composer, with options to ignore GD extension requirements. ### Tech Stack: PHP ### Why It Matters: LLPhant simplifies the integration of Generative AI and vector databases into PHP applications, making it easier to build powerful and efficient AI-driven solutions. Its compatibility with popular frameworks like Symfony and Laravel, combined with its support for various LLMs, makes it a valuable tool for developers looking to leverage cutting-edge AI technologies in their projects.
Go 1,721 stars

aoci-spec/aoci-code

A persistent, Git-versioned map of your whole codebase and database schema that coding agents read before they touch anything. Local-first MCP server + CLI in Go: a governed repository index of code knowledge that gives Claude Code, Codex, Cursor, and opencode long-term context, memory, and code intelligence.

### Overview: AOCI-CODE is a persistent, Git-versioned map of your entire codebase and database schema. It provides context and intelligence to coding agents, ensuring they know the system before starting tasks. This local-first MCP server and CLI in Go gives agents long-term context, memory, and code intelligence. ### Key Features: - **Persistent and Versioned Map**: Tracks your entire codebase and database schema. - **Contextual Knowledge**: Provides context and intelligence to coding agents. - **Local First MCP Server**: Governs the repository index, ensuring agents know the system before starting tasks. - **CLI Interface**: Command-line interface for easy integration. - **Governing Index**: A repository index versioned by Git, ensuring the index matches the current code. ### Tech Stack: Go ### Why It Matters: - Ensures coding agents have the necessary context and intelligence before starting tasks. - Facilitates efficient development by reducing the need for re-reading the codebase. - Enables seamless transitions between projects and agents, maintaining the index across changes. - Automates maintenance of the index, ensuring it always matches the current codebase. Repository: aoci-spec/aoci-code Primary Language: Go GitHub Stars: 1699
LLVM 1,720 stars

EnzymeAD/Enzyme

High-performance automatic differentiation of LLVM and MLIR.

Overview: Enzyme is a high-performance automatic differentiation tool that optimizes the performance of statically analyzable LLVM and MLIR code, surpassing state-of-the-art AD tools. Key features: - Automatic differentiation of LLVM and MLIR code - Efficiently performs AD on optimized code - Supports GPU optimizations and parallel program AD - Compatible with various programming languages and frameworks Tech stack: LLVM Why it matters: Enzyme provides a powerful and efficient solution for automatic differentiation, making it easier and faster to perform AD on complex systems, especially those involving multiple parallel paradigms and parallel programming languages like Julia, OpenMP, MPI, and Rust. Repository: EnzymeAD/Enzyme GitHub stars: 1720 GitHub description: High-performance automatic differentiation of LLVM and MLIR.
Jupyter Notebook 1,720 stars

neomatrix369/awesome-ai-ml-dl

Awesome Artificial Intelligence, Machine Learning and Deep Learning as we learn it. Study notes and a curated list of awesome resources of such topics.

### Overview: This repository is a curated list of awesome resources for Artificial Intelligence, Machine Learning, and Deep Learning. It includes guides, domains, tools, and examples to help you learn and explore these topics. ### Key Features: - Python and Data basics - Core topics: ML, DL, NLP - Do stuff: Notebooks, Examples, Tools ### Tech Stack: Jupyter Notebook ### Why it Matters: Explore and learn about AI/ML/DL with curated guides, domains, tools, and examples. ### Repository: Repository: neomatrix369/awesome-ai-ml-dl GitHub Stars: 1720 Repository Description: Awesome Artificial Intelligence, Machine Learning and Deep Learning as we learn it. Study notes and a curated list of awesome resources of such topics. ### Table of Contents: - What's new - Start here - Explore by domain - AI Agents - NLP - Computer Vision - LLMs - Generative AI - MLOps & Deployment - Time Series & Anomaly Detection ### Additional Resources: - [ContinueWith](https://continuewith.ai) - Let visitors continue any website page inside ChatGPT, Claude, Gemini, Grok, Perplexity, Mistral, and other AI assistants in one click. ### Sponsoring: If you find this repository useful, consider sponsoring it to support its maintenance and growth. ### Disclaimer: This repository is maintained by a community of contributors. The information provided is based on the available
Python 1,719 stars

lotus-data/lotus

Optimized Agentic and LLM Bulk Processing Over Your Data

### Overview: LOTUS is an optimized bulk processing framework designed for both agentic and language model (LLM) tasks. It leverages semantic operators to enhance the accuracy and cost-effectiveness of processing large datasets with LLMs and natural language instructions. LOTUS optimizes these operations to provide higher accuracy and lower costs. ### Key Features: - **Agentic Code Processing**: Run a tool-using agent over every file, document, or record, then reduce to one answer. - **Deep Research & Synthesis**: Fan out over a corpus, extract, and synthesize. - **Agent-Trace Failure Analysis**: Mine large volumes of agent logs for failure modes. - **Document Extraction & Unstructured Analysis**: Extract structured fields and insights from text. - **LLM-Judge Evaluations**: Score model outputs, app responses, and content quality, row by row. - **RAG**: Retrieve the relevant context and generate grounded answers over your data. ### Tech Stack: Python ### Why It Matters: LOTUS optimizes LLM-based operations to enhance accuracy and reduce costs, making it ideal for large-scale data processing tasks. It supports various data types and provides a robust framework for both agentic and language model tasks. ### Repository: - **Repository**: lotus-data/lotus - **GitHub Stars**: 1719 ### Primary Language: - **Primary Language**: Python ### Installation: ```bash pip install lotus-ai ``` ### Community: - **Discord**: Join the community on Discord (https://discord.gg/join) -
Python 1,719 stars

tenstorrent/tt-metal

:metal: TT-NN operator library, and TT-Metalium low level kernel programming model.

### Overview: TT-NN, the TT-Metalium low-level kernel programming model, is a comprehensive Python & C++ neural network operator library designed for high-performance, low-latency AI inference. It offers optimized performance, compatibility across various hardware platforms, and seamless integration with the vLLM inference server. ### Key Features: - **TT-NN Library**: A powerful Python & C++ neural network operator library. - **TT-Metalium Kernel**: A low-level kernel programming model for efficient hardware acceleration. - **Performance Metrics**: Measures include Time to First Token (TTFT), Tokens per Second per User (T/S/U), Tokens per Second (T/S), Tensor Parallel (TP), and Data Parallel (DP) performance. - **Model Demos**: Provides model demonstrations for various hardware configurations, including Llama 3.3 70B, Qwen 2.5 7B, and Qwen 2.5 72B. ### Tech Stack: - Python - C++ ### Why It Matters: TT-NN and TT-Metalium offer unparalleled performance, low latency, and compatibility across a wide range of hardware platforms, making them ideal for real-time AI applications and edge computing environments. The library's optimized design ensures efficient resource utilization, leading to significant improvements in inference speed and accuracy. Repository: tenstorrent/tt-metal Primary Language: Python GitHub Stars: 1717 GitHub Description: :metal: TT-NN operator library, and TT-Metalium low level kernel programming model.
Jupyter Notebook 1,718 stars

R6410418/Jackrong-llm-finetuning-guide

### Overview: This repository is a comprehensive educational resource portal designed for beginners and developers. It covers a wide range of topics including fine-tuning, dataset distillation, reinforcement learning, and local deployment of large language models like Qwen. The repository provides reproducible training pipelines, SFT and RL workflows, data preparation and distillation recipes, 16-bit export and GGUF deployment workflows, and agent-ready Qwen MTP GGUF conversion tools. ### Key Features: - Fine-tuning of LLM models - Dataset distillation - Reinforcement learning workflows - Local deployment of models - 16-bit export and GGUF deployment - Agent-ready Qwen MTP GGUF conversion ### Tech Stack: Jupyter Notebook ### Why it Matters: This repository serves as a valuable resource for anyone looking to understand and implement advanced techniques in large language model training and deployment, making it easier for developers to create and maintain their own models. ### Repository: R6410418/Jackrong-llm-finetuning-guide ### Primary Language: Jupyter Notebook ### GitHub Stars: 1716
Python 1,718 stars

allenai/scibert

A BERT model for scientific text.

Overview: Scibert is a pre-trained contextualized embedding model designed for scientific text analysis, including named entity recognition, relation extraction, participant intervention comparison outcome, and more. Key features: - Named entity recognition - Relation extraction - Participant intervention comparison outcome - Named entity recognition for NCBIDisease - Sentence classification for paper fields - Citation intent classification - Sentence classification for ACL-ARC - Dependency parsing for Genia-Las and Genia-UAS - Named entity recognition for JNLPA Tech stack: Python Why it matters: Scibert provides a powerful toolset for scientific text analysis, enabling researchers to extract meaningful information from complex scientific documents with high accuracy and efficiency.
Rust 1,718 stars

fynnfluegge/agtx

🏄🏼‍♂️ The blackboard for coding agents - agentic development environment for claude code, codex, cursor, opencode, grok and more.

Overview: agtx is a terminal-native agentic development environment designed for 10x productivity, providing a shared board where multiple coding agents can collaborate on the same task. It supports agents like Claude Code, Codex, Grok, Cursor, OpenCode, Antigravity, Gemini CLI, and Copilot, allowing for parallel task lifecycles and automatic agent switching. Key features: - **Supported Agents**: Claude Code, Codex, Grok, Cursor, OpenCode, Antigravity, Gemini CLI, Copilot, and pi. - **Parallel Multi-agent Task Lifecycle**: Configure different agents per workflow phase, e.g., Grok for research, Claude for implementation, Codex for review, and switch agents automatically. - **Multi-project Kanban Board**: Manage agent sessions across all projects using a single terminal-based TUI without leaving the terminal. - **Vim-native Keybindings**: Control the board and agent sessions with Vim-powered muscle memory. - **Mobile Support**: Press `W` to scan a QR code for an install. Tech stack: Rust Why it matters: agtx simplifies the complex process of coding agents by providing a unified environment for multiple agents to work together, enhancing productivity and collaboration in the development of AI models.
Rust 1,717 stars

gi-dellav/zerostack

Lightweight coding agent written in Rust, optimized for memory footprint and performance

### Overview: A lightweight and efficient coding agent written in Rust, optimized for memory footprint and performance. ### Key features: - **Multi-provider**: Supports multiple open-source providers. - **Standard tools**: Exposes all standard tools described by opencode documentation. - **Permission system**: Configurable modes with session allowlists and policies. - **Session management**: Saves, loads, and resumes sessions. - **Terminal UI**: CROSSTERM-based, supports markdown rendering, mouse selection, and scrollback. - **Prompts system**: Switches between different prompt modes. - **MCP support**: Optional compile-time feature for extended tooling. - **Integrated Exa search**: Enables WebFetch and WebSearch tools. - **Integrated Ralph Wiggum loops**: Supports looping capabilities. - **Integrated Git Worktrees integration**: Allows moving the agent between worktrees. - **ACP support**: Agent Communication Protocol server for editor connections. - **Persistent memory**: A global MEMORY.md and daily logs injected into the system prompt. - **Lifecycle hooks**: Observes or gates tool calls, prompts, and session lifecycle events. - **Advisor**: A second model for strategic guidance, with a human-handoff mode. - **Subagents**: Parallel and fast, useful for exploring the codebase. - **ARCHITECTURE.md**: Companion file for shared core knowledge. - **Prompt chaining**: Advanced task chaining for brainstorming, planning, coding, and reviewing. - **Status signals**: Emit run-state events over a Unix socket for external status bars. ### Tech stack: Rust. ### Why
Rust 1,716 stars

AnalyseDeCircuit/oxideterm

AI-native workspace for local shells and remote machines.Zero Webview, zero OpenSSL, zero telemetry, and no app subscription.Written in Rust+GPUI

### Overview: OxideTerm is a powerful, AI-native SSH client and remote operations workspace designed for local shells and remote machines. It offers a zero-Telemetry, zero-Webview, and zero-subscription experience, built entirely in Rust with a GPU-rendered app. ### Key Features: - **SSH, Mosh, Telnet, Serial, RDP/VNC, SFTP, Port Forwarding, Built-in Editor, Terminals, Protocols, tmux, Broadcast, Reliability, Files, Editing, Networking, Remote Desktop, Host Operations, AI, Automation, Review, Audit, Sync, Portability, Personalization** - **No Account, No Subscription, No Telemetry, No Electron** - **GPU-Rendered App, Zero Webview, Zero OpenSSL, Zero Telemetry, Zero App Subscription** ### Tech Stack: Rust + GPUI ### Why It Matters: OxideTerm provides a seamless, secure, and efficient way to manage remote operations, with AI-assisted features for a more personalized experience. It eliminates the need for an account, subscription, or telemetry, making it ideal for developers and teams looking for a lightweight, zero-impact solution.
Python 1,716 stars

lightly-ai/lightly-train

All-in-one training for vision models (YOLO, ViTs, RT-DETR, DINOv3): pretraining, fine-tuning, distillation.

### Overview: LightlyTrain is a comprehensive framework designed to accelerate the development and deployment of state-of-the-art computer vision models. It supports pretraining, fine-tuning, and distillation across various vision tasks, including object detection, semantic segmentation, and more. ### Key Features: - **Pretraining**: LightlyTrain pretrains models on large datasets like COCO, enabling efficient fine-tuning. - **Fine-tuning**: Supports fine-tuning of YOLO and ViTs models on detection and segmentation tasks. - **Distillation**: Offers distillation methods to improve generalization across different tasks and models. - **Tiny Models**: Provides tiny DINOv3 models for low-power embedded devices. - **Edge Deployment**: Supports ONNX/TensorRT export for faster inference on edge devices. ### Tech Stack: Python ### Why It Matters: LightlyTrain simplifies the entire model development lifecycle, from pretraining to fine-tuning, making it an essential tool for researchers and developers looking to achieve high performance and efficiency in computer vision projects.
Python 1,716 stars

scverse/scvi-tools

Deep probabilistic analysis of single-cell and spatial omics data

Overview: scvi-tools is a comprehensive library for probabilistic analysis of single-cell omics data, built on PyTorch and AnnData. It includes models for dimensionality reduction, data integration, annotation, factor analysis, and more. Key features: - Dimensionality reduction - Data integration - Automated annotation - Factor analysis - Doublet detection - Spatial deconvolution Tech stack: Python, PyTorch, Pyro, AnnData Why it matters: scvi-tools provides a powerful toolkit for analyzing and interpreting single-cell omics data, making it essential for researchers in the field. Repository: scverse/scvi-tools GitHub stars: 1716 GitHub description: Deep probabilistic analysis of single-cell and spatial omics data [![Stars][gh-stars-badge]][gh-stars-link] [![PyPI][pypi-badge]][pypi-link] [![PyPIDownloads][pepy-badge]][pepy-link] [![CondaDownloads][conda-badge]][conda-link] [![Docs][docs-badge]][docs-link] [![Tests][tests-badge]][tests-link] [![Build][build-badge]][build-link] [![Coverage][coverage-badge]][coverage-link] [scvi-tools] (single-cell variational inference tools) is a package for probabilistic modeling and analysis of single-cell omics data, built on top of [PyTorch] and [AnnData]. # Analysis of single-cell omics data scvi-tools is composed of models that perform many analysis tasks across single-cell, multi, and
Unknown 1,714 stars

vahidk/EffectivePyTorch

PyTorch tutorials and best practices.

Overview: This repository contains tutorials and best practices for effective use of PyTorch, a popular library for numerical computation and machine learning research. Key features: - PyTorch basics - Encapsulate your model with Modules - Broadcasting the good and the ugly - Take advantage of the overloaded operators - Optimize runtime with TorchScript - Build efficient custom data loaders - Numerical stability in PyTorch - Faster training with automatic mixed precision Tech stack: Python Why it matters: PyTorch is widely used in machine learning research and is known for its simplicity and efficiency. It allows for easy integration of GPUs and TPUs, making it highly suitable for training large models. The tutorials and best practices provided in this repository aim to help developers optimize their PyTorch projects and achieve better performance.
Python 1,714 stars

xingyaoww/code-act

Official Repo for ICML 2024 paper "Executable Code Actions Elicit Better LLM Agents" by Xingyao Wang, Yangyi Chen, Lifan Yuan, Yizhe Zhang, Yunzhu Li, Hao Peng, Heng Ji.

### Overview: This project introduces **CodeAct**, a novel approach to consolidate and execute LLM agent actions into a unified action space. By integrating a Python interpreter, CodeAct allows LLM agents to perform code actions and dynamically adapt based on new observations. This method significantly improves the performance and adaptability of LLM agents compared to traditional textual approaches. ### Key Features: - **CodeAct Agent**: A unified action space for LLM agents that can execute code actions and dynamically revise actions based on new observations. - **CodeActInstruct**: A dataset of 7k multi-turn interactions using CodeAct, collected for training and evaluation. - **CodeActAgent-Mistral-7b-v0.1**: A variant of CodeActAgent trained on CodeActInstruct, using Mistral-7b-v0.1 as the base model with 32k context window. - **CodeActAgent-Llama-7b**: A variant trained on CodeActInstruct, using Llama-2-7b as the base model with 4k context window. ### Tech Stack: Python ### Why It Matters: Our extensive analysis of 17 LLMs on API-Bank and a newly curated benchmark [M<sup>3</sup>ToolEval](docs/EVALUATION.md) shows that CodeAct outperforms widely used alternatives like Text and JSON (up to 20% higher success rate). Please check our paper for more detailed analysis!