GTP engine and self-play learning in Go

Autoresearch for Go

Modern RL Post-training Infrastructure: Optimized for NVIDIA/AMD GPUs with a focus on vLLM and DeepSpeed integration, CUDA/ROCm/Triton kernels, and transparent hardware-aware scaling.

Graphic notes on Gilbert Strang's "Linear Algebra for Everyone"

Cosmos-RL is a flexible and scalable Reinforcement Learning framework specialized for Physical AI applications.

A programmable distributed training system for PyTorch

Research artifacts from Recursive's automated AI research system

A profiling and performance analysis tool for machine learning

An LLM post-training framework with vLLM for RL Scaling

A unified framework for building, running, and training general agents at scale.

UniRL is a Framework for Unified Multimodal Model Reinforcement Learning

Compositional Muon release

Official implementation of GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization

torch_remat fine-grained activation checkpointing API

Provide performance insight capabilities for RL frameworks.

OpenClaw-RL: Train any agent simply by talking

Orchestrate multiple coding agents from desktop and mobile

GPU-accelerated LLM Training Simulator

🔥 LeetCode for PyTorch — practice implementing softmax, attention, GPT-2 and more from scratch with instant auto-grading. Jupyter-based, self-hosted or try online.

CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs

A high-performance RL training-inference weight synchronization framework, designed to enable second-level parameter updates from training to inference in RL workflows

🎨 NeMo Data Designer: Generate high-quality synthetic data from scratch or from seed data.

A Lightweight LLM Post-Training Library

An asynchronous streaming data management module for efficient post-training.

Pipeline parallelism for the minimalist

Artefacts from the first complete run of the Lossfunk AI Scientist pipeline for paper accepted at Agents4Science 2025.