165 repositories starred since Apr 2010. Mostly machine learning and data, and lately AI and language models.

Run the full 2.78-trillion-parameter Kimi K3 model, DeepSeek V4.1 Flash or GLM-5.3-Flash beyond available RAM by streaming activated weights directly from NVMe. A dependency-free, embeddable C inference engine.

Multi-subscription extension for pi -- use multiple OAuth accounts per provider (Anthropic, Codex, Copilot, Gemini, Antigravity)

Very low latency speech to text, intent recognition, and text to speech, for building voice agents and interfaces

Minimal Claude Code alternative. Single Python file, zero dependencies, ~250 lines.

Our library for RL environments + evals

FlexAttention based, minimal vllm-style inference engine for fast Gemma 2 inference.

gpt-oss-120b and gpt-oss-20b are two open-weight language models by OpenAI

Wan: Open and Advanced Large-Scale Video Generative Models

The simplest, fastest repository for training/finetuning small-sized VLMs.

Implementing DeepSeek R1's GRPO algorithm from scratch

Frontier Models playing the board game Diplomacy.

Minimal reproduction of DeepSeek R1-Zero

Official repository for our work on micro-budget training of large-scale diffusion models.

NanoGPT (124M) in 90 seconds

A PyTorch native platform for training generative AI models

Efficient Triton Kernels for LLM Training

Apple MLX native implementations of state-of-the-art generative image & video models

Official inference repo for FLUX.1 models

Run PyTorch LLMs locally on servers, desktop and mobile

Simple Byte pair Encoding mechanism used for tokenization process . written purely in C

UNet diffusion model in pure CUDA

[ICLR 2025] Samba: Simple Hybrid State Space Models for Efficient Unlimited Context Language Modeling

gpt-2 from scratch in mlx

User-friendly AI Interface (Supports Ollama, OpenAI API, ...)

Implementation for MatMul-free LM.

Vane is an AI-powered answering engine.

LLM inference in C/C++

Distribute and run LLMs with a single file.

Fast bare-bones BPE for modern tokenizer training

A benchmark to evaluate language models on questions I've previously asked them to solve.

Official implementation for the paper: "Code Generation with AlphaCodium: From Prompt Engineering to Flow Engineering""

A high-throughput and memory-efficient inference and serving engine for LLMs

RuLES: a benchmark for evaluating rule-following in language models

Fine-tune mistral-7B on 3090s, a100s, h100s

Simple and efficient pytorch-native transformer text generation in <1000 LOC of python.

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

A python script to help manage a Gmail inbox by filtering out promotional emails using GPT-3 or GPT-4.

20+ high-performance LLMs with recipes to pretrain, finetune and deploy at scale.

The TinyLlama project is an open endeavor to pretrain a 1.1B Llama model on 3 trillion tokens.

Implementation of the LLaMA language model based on nanoGPT. Supports flash attention, Int8 and GPTQ 4bit quantization, LoRA and LLaMA-Adapter fine-tuning, pre-training. Apache 2.0-licensed.

OpenAssistant is a chat-based assistant that understands tasks, can interact with third-party systems, and retrieve information dynamically to do so.

The agent engineering platform.

LlamaIndex is the document processing platform for AI

Hackable and optimized Transformers building blocks, supporting a composable construction.

Robust Speech Recognition via Large-Scale Weak Supervision

🤗 Diffusers: State-of-the-art diffusion models for image, video, and audio generation in PyTorch.

Official implementation of VQ-Diffusion

DALL·E Mini - Generate images from a text prompt

Taming Transformers for High-Resolution Image Synthesis