Sparse autoencoders

Tile primitives for speedy kernels

A fast TCP/UDP tunnel over HTTP

Delightful io_uring packages and resources

Universal Tensor Operations in Einstein-Inspired Notation for Python.

tiktoken is a fast BPE tokeniser for use with OpenAI's models.

A simple, performant and scalable Jax LLM!

NCCL Tests

Swarm training framework using Haiku + JAX + Ray for layer parallel transformer language models on unreliable, heterogeneous nodes

🏛️A research-friendly codebase for fast experimentation of single-agent reinforcement learning in JAX • End-to-End JAX RL

Deep learning at the speed of light.

GPU & Accelerator process monitoring for AMD, Apple, Huawei, Intel, NVIDIA and Qualcomm

Analysis of various tricky Rust code

CUDA related news and material links

Multi-Agent Reinforcement Learning with JAX

magic-trace collects and displays high-resolution traces of what a process is doing

Implementation of Q-Transformer, Scalable Offline Reinforcement Learning via Autoregressive Q-Functions, out of Google Deepmind

A lightweight, memory-protected, message-passing kernel for deeply embedded systems.

Type annotations and runtime checking for shape and dtype of JAX/NumPy/PyTorch/etc. arrays. https://docs.kidger.site/jaxtyping/

Elegant easy-to-use neural networks + scientific computing in JAX. https://docs.kidger.site/equinox/

You like pytorch? You like micrograd? You love tinygrad! ❤️

ctypes wrappers for HIP, CUDA, and OpenCL

MLX: An array framework for Apple silicon

Massively parallel rigidbody physics simulation on accelerator hardware.

A self contained OBS -> FTL -> WebRTC live streaming server. Comprised of 3 parts once configured anyone can achieve sub-second OBS to the browser livestreaming

This repository contains implementations and illustrative code to accompany DeepMind publications

Flask-based package for monitoring utilisation of nVidia GPUs.