PyTorch-native post-training at scale

(best/better) practices of megatron on veRL and tuning guide

RLinf is a flexible and scalable open-source infrastructure designed for post-training foundation models (LLMs, VLMs, VLAs) via reinforcement learning.

Verlog: A Multi-turn RL framework for LLM agents

Official repo for IRL-VLA

Chain-of-Agents: End-to-End Agent Foundation Models via Multi-Agent Distillation and Agentic RL.

MiroRL is an MCP-first reinforcement learning framework for deep research agent.

JAxtar is a project with a JAX-native implementation of parallelizeable A* & Q* solver for neural heuristic search research.

GRPO training code which scales to 32xH100s for long horizon terminal/coding tasks. Base agent is now the top Qwen3 agent on Stanford's TerminalBench leaderboard.

vsag is a vector indexing library used for similarity search.

Simple and efficient DeepSeek V3 SFT using pipeline parallel and expert parallel, with both FP8 and BF16 trainings

The 100 line AI agent that solves GitHub issues or helps you in your command line. Radically simple, no crazy configs, no giant monorepo—but scores 65% on SWE-bench verified!

Lightweight and portable LLM sandbox runtime (code interpreter) Python library.

Hierarchical Reasoning Model Official Release

My learning notes/codes for ML SYS.

🔥🔥🔥Latest Papers, Codes on Uncertainty-based RL

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

TreeRL: LLM Reinforcement Learning with On-Policy Tree Search in ACL'25

[ACL 2025] This is the official implementation for the paper: "Meta-rater: A Multi-dimensional Data Selection Method for Pre-training Language Models"

A library for advanced large language model reasoning