MUSA Templates for Linear Algebra Subroutines

Production ready LLM model compression/quantization toolkit with accelerated inference support for both cpu/gpu via HF, vLLM, and SGLang.

Amnezia VPN Client (Desktop+Mobile)

Fast Hadamard transform in CUDA, with a PyTorch interface

Virtual environment stacks for Python

NanoGPT (124M) quality in 2.4B tokens

LLM training in simple, raw C/CUDA

MobileLLM Optimizing Sub-billion Parameter Language Models for On-Device Use Cases. In ICML 2024.

Fast and memory-efficient exact attention

Meta Lingua: a lean, efficient, and easy-to-hack codebase to research LLMs.

Quantized Attention that achieves speedups of 2.1x and 2.7x compared to FlashAttention2 and xformers, respectively, without lossing end-to-end metrics across various models.

The code and data for "MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark" [NeurIPS 2024]

Official inference framework for 1-bit LLMs

Convert PDF to markdown quickly with high accuracy

EfficientQAT: Efficient Quantization-Aware Training for Large Language Models

Tile primitives for speedy kernels

Repo for "LoLCATs: On Low-Rank Linearizing of Large Language Models"

llama3 implementation one matrix multiplication at a time

Code for the paper "Entanglement-induced provable and robust quantum learning advantages"

Entropy Based Sampling and Parallel CoT Decoding

VPTQ, A Flexible and Extreme low-bit quantization algorithm

Powerful system container and virtual machine manager

A MAD laboratory to improve AI architecture designs 🧪

A dungeon crawler designed for a quantum computer

A port of DOOM for a quantum computer

StutterFormer is an AI model that aims to be able to receive a speech sample with stuttering disfluencies, and return it with the disfluencies attenuated or eliminated.

Filament is a real-time physically based rendering engine for Android, iOS, Windows, Linux, macOS, and WebGL2