Nano vLLM

Build RL environments for LLM training

A PyTorch native platform for training generative AI models

A scalable asynchronous reinforcement learning implementation with in-flight weight updates.

OLMoE: Open Mixture-of-Experts Language Models

My learning notes for ML SYS.

Post-training with Tinker

This repository contains the official implementation of "FastVLM: Efficient Vision Encoding for Vision Language Models" - CVPR 2025

Tongyi Deep Research, the Leading Open-source Deep Research Agent

VS Code extension for Pyroscope continuous profiling with inline metrics and heat maps

Thermodynamic Hypergraphical Model Library in JAX

PyTorch-native post-training at scale

Expert Parallelism Load Balancer

A Python-embedded DSL that makes it easy to write fast, scalable ML kernels with minimal boilerplate.

Contexts Optical Compression

Fastest kernels written from scratch

StreamingVLM: Real-Time Understanding for Infinite Video Streams

The best ChatGPT that $100 can buy.

A novel data compression framework

Perplexity GPU Kernels

Simple high-throughput inference library

A Lightweight LLM Post-Training Library

A Datacenter Scale Distributed Inference Serving Framework

Mooncake is the serving platform for Kimi, a leading LLM service provided by Moonshot AI.

verl: Volcano Engine Reinforcement Learning for LLMs

Language modeling with linear-cost context

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

Ongoing research training transformer models at scale

The Meta Spatial SDK Samples is a collection of code samples and projects that demonstrate the capabilities of the Meta Spatial SDK. Meta Spatial SDK enables mobile developers to build Meta Horizon…