Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels

Optimize prompts, code, and more with AI-powered Reflective Text Evolution

Puffing up reinforcement learning

Draco is a library for compressing and decompressing 3D geometric meshes and point clouds. It is intended to improve the storage and transmission of 3D graphics.

Continual reinforcement learning recipes for LLMs.

The best-benchmarked open-source AI memory system. And it's free.

A light-weight and powerful meta-prompting, context engineering and spec-driven development system for Claude Code by TÂCHES.

Distributed Task Queue (development branch)

Ongoing research training transformer models at scale

Best practices for training DeepSeek, Mixtral, Qwen and other MoE models using Megatron Core.

[CSUR 2025] Continual Learning of Large Language Models: A Comprehensive Survey

Karpathy-style autoresearch for Mamba language models on MLX and GPU.

Train the smallest LM you can that fits in 16MB. Best model wins!

Post-training with Tinker

A Simple and Universal Swarm Intelligence Engine, Predicting Anything. 简洁通用的群体智能引擎，预测万物

Official inference framework for 1-bit LLMs

An interoperable Python framework for biomolecular simulation.

AI agents running research on single-GPU nanochat training automatically

Hypernetworks that update LLMs to remember factual information

JSON for Modern C++

Memory augmented inference library for Recursive Language Models (RLMs), built on top of rlm.

General plug-and-play inference library for Recursive Language Models (RLMs), supporting various sandboxes.

Agent2Agent (A2A) is an open protocol enabling communication and interoperability between opaque agentic applications.

verl/HybridFlow: A Flexible and Efficient RL Post-Training Framework

FAIR Chemistry's library of machine learning methods for chemistry

A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.

STOKE: A stochastic superoptimizer and program synthesizer

A tiling window manager for macOS based on binary space partitioning

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