A modular, primitive-first, python-first PyTorch library for Reinforcement Learning.

A Python-embedded DSL that makes it easy to write fast, scalable ML kernels with minimal boilerplate.

Alerting system for the pytorch org

Collective communications library with various primitives for multi-machine training.

Fault tolerance for PyTorch (HSDP, LocalSGD, DiLoCo, Streaming DiLoCo)