A Next-Generation Training Engine Built for Ultra-Large MoE Models

I recently interviewed with some AI labs and these are the notes I took during my study for ML fundamentals and Design. This was in Mar 2025 and given how fast the field of AI moves, some of it may…

Research prototype of PRISM — a cost-efficient multi-LLM serving system with flexible time- and space-based GPU sharing.

Materials for learning SGLang

Renderer for the harmony response format to be used with gpt-oss

A Distributed Attention Towards Linear Scalability for Ultra-Long Context, Heterogeneous Data Training

Allow torch tensor memory to be released and resumed later

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

Long-RL: Scaling RL to Long Sequences

slime is a LLM post-training framework aiming at scaling RL.

The official repo of Pai-Megatron-Patch for LLM & VLM large scale training developed by Alibaba Cloud.

Ongoing research training transformer models at scale

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

Bridge Megatron-Core to Hugging Face/Reinforcement Learning

SkyRL: A Modular Full-stack RL Library for LLMs

Github Pages template for academic personal websites, forked from mmistakes/minimal-mistakes

Code to automatically prove or verify estimates in analysis

Allow torch tensor memory to be released and resumed later

My learning notes/codes for ML SYS.

Some Environment Examples of LLM Agents, it's designed to be able to integrated with VeRL

Distributed RL System for LLM Reasoning

Agent Laboratory is an end-to-end autonomous research workflow meant to assist you as the human researcher toward implementing your research ideas

My learning notes/codes for ML SYS.

Production-tested AI infrastructure tools for efficient AGI development and community-driven innovation

verl: Volcano Engine Reinforcement Learning for LLMs

Fully open reproduction of DeepSeek-R1

Sky-T1: Train your own O1 preview model within $450