Autoresearch for GPU kernels. Give it any PyTorch model, go to sleep, wake up to optimized Triton kernels.

A debugging and profiling tool that can trace and visualize python code execution

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

RL post-training open LLMs for math reasoning

Minimal library for distributed python work. Can efficiently run CPU and GPU tasks across 100s of machines.

Glommio is a thread-per-core crate that makes writing highly parallel asynchronous applications in a thread-per-core architecture easier for rustaceans.

Find dead Python code

RedKit is a lightweight, high-performance Redis-compatible server framework written in Go

Sharp Monocular View Synthesis in Less Than a Second

Async RL Training at Scale

Miles is an enterprise-facing reinforcement learning framework for LLM and VLM post-training, forked from and co-evolving with slime.

Example for implementing isolated environments with Ray and UV

High-quality single file implementation of Deep Reinforcement Learning algorithms with research-friendly features (PPO, DQN, C51, DDPG, TD3, SAC, PPG)

PyTorch-native post-training at scale

🔬 A fast, interactive web-based viewer for performance profiles.

A modern replacement for Redis and Memcached

The best ChatGPT that $100 can buy.

Official PyTorch implementation for "Large Language Diffusion Models"

A natively parallel dataloader for Python, written in Rust. Serving data at GB/s speeds, while covering aspect ratio bucketing, crop and resize for image ML workloads.

A fast Rust JSON library based on SIMD.

DeepEP: an efficient expert-parallel communication library

The `io_uring` library for Rust

The source code for the blog post The 37 Implementation Details of Proximal Policy Optimization

H-Net: Hierarchical Network with Dynamic Chunking

Kimi K2 is the large language model series developed by Moonshot AI team

Simplifying reinforcement learning for complex game environments

PyTorch Single Controller

Mirage Persistent Kernel: Compiling LLMs into a MegaKernel