Stars
MoE training for Me and You and maybe other people
Fault tolerance for PyTorch (HSDP, LocalSGD, DiLoCo, Streaming DiLoCo)
A fancy self-hosted monitoring tool
A fast GPU memory copy library based on NVIDIA GPUDirect RDMA technology
RoboCasa: Large-Scale Simulation of Everyday Tasks for Generalist Robots
MiniMax-M2, a model built for Max coding & agentic workflows.
A Datacenter Scale Distributed Inference Serving Framework
Qwen3-omni is a natively end-to-end, omni-modal LLM developed by the Qwen team at Alibaba Cloud, capable of understanding text, audio, images, and video, as well as generating speech in real time.
SQL databases in Python, designed for simplicity, compatibility, and robustness.
Official inference repo for FLUX.1 models
Post-training with Tinker
DeepEP: an efficient expert-parallel communication library
Code implementation for the paper "Large-scale Pre-training for Grounded Video Caption Generation" (ICCV 2025)
An extremely fast Python package and project manager, written in Rust.
Development repository for the Triton language and compiler
Renderer for the harmony response format to be used with gpt-oss
DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling
High-Resolution 3D Assets Generation with Large Scale Hunyuan3D Diffusion Models.
A PyTorch native platform for training generative AI models
MCP server that enables AI assistants to interact with Linear project management system through natural language, allowing users to retrieve, create, and update issues, projects, and teams.
QAlign is a new test-time alignment approach that improves language model performance by using Markov chain Monte Carlo methods.