A high-performance pipeline-parallel runtime for large-scale MoE serving of agentic workloads. Also, a concise inference engine for curriculum and research purpose.

mKernel: fast multi-node, multi-GPU fused kernels

Easy-to-Use Distributed Inference API for PyTorch

A distributed training framework based on torchtitan

A unified, high-performance framework for training LLMs, VLMs, diffusion, and embodied models on NVIDIA GPUs and Kunlun XPUs.

Embodied AI Operating System (EAIOS)

Skeleton code for CPP version raft lab

PhyAI is a high-performance framework for running Physical AI models (VLA, WAM, and beyond), supporting both cloud-based serving and on-device deployment.

DeepSpec: a full-stack codebase for training and evaluating speculative decoding algorithms

A unified workflow orchestration and distributed KV cache sharing framework for multimodal inference.

The high-performance distributed tensor layer — load once, share everywhere.

Frontier: A Discrete-Event Simulator for Modern LLM Serving

Autonomous CUDA kernel optimization agent with structured task specs and per-config scoring

Stable and Efficient Reinforcement Learning for Trillion-Parameter LLMs

An asynchronous streaming data management module for efficient post-training.

The Language Virtual Machine for Agent Skills

TokenSpeed is a speed-of-light LLM inference engine.

A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.

21 writing rules for AI coding and writing agents. Drop-in for Claude Code, Codex, Copilot, Cursor, and Aider, so their output reads like a tech pro.