OpenClaw-RL: Train any agent simply by talking

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

slime is an LLM post-training framework for RL Scaling.

Bash is all you need - A nano claude code–like 「agent harness」, built from 0 to 1

DFlash: Block Diffusion for Flash Speculative Decoding

Official implementation of "FOCUS: DLLMs Know How to Tame Their Compute Bound".

Virtualized Elastic KV Cache for Dynamic GPU Sharing and Beyond

cuTile is a programming model for writing parallel kernels for NVIDIA GPUs

torchcomms: a modern PyTorch communications API

The best ChatGPT that $100 can buy.

Fast low-bit matmul kernels in Triton

Pipeline parallelism for the minimalist

[MobiCom'25] Elastic On-Device LLM Service.

An Easy-to-use, Scalable and High-performance Agentic RL Framework based on Ray (PPO & DAPO & REINFORCE++ & VLM & TIS & vLLM & Ray & Async RL)

Allow torch tensor memory to be released and resumed later

Official implementation for DenseMixer: Improving MoE Post-Training with Precise Router Gradient

extensible collectives library in triton

📚 Modern C++ Tutorial: C++11/14/17/20 On the Fly | https://changkun.de/modern-cpp/

Distributed Compiler based on Triton for Parallel Systems

A PyTorch native platform for training generative AI models

verl: Volcano Engine Reinforcement Learning for LLMs

Tile primitives for speedy kernels

CUDA Python: Performance meets Productivity

Mamba SSM architecture

[ASPLOS'25] Towards End-to-End Optimization of LLM-based Applications with Ayo

[SenSys'24] PieBridge: Fast and Parameter-Efficient On-Device Training via Proxy Networks

DeepEP: an efficient expert-parallel communication library

Official Repo for Open-Reasoner-Zero