Playing around "Less Slow" coding practices in C++ 20, C, CUDA, PTX, & Assembly, from numerics & SIMD to coroutines, ranges, exception handling, networking and user-space IO

Tile primitives for speedy kernels

Fast low-bit matmul kernels in Triton

lsblk in go for apple computers

A categorized list of C++ resources.

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

CUDA on non-NVIDIA GPUs

SPIRV-Cross is a practical tool and library for performing reflection on SPIR-V and disassembling SPIR-V back to high level languages.

Any model. Any hardware. Zero compromise. Built with @ziglang / @openxla / MLIR / @bazelbuild

Whisper with Medusa heads

A high-performance, zero-overhead, extensible Python compiler with built-in NumPy support

convert images, video to ascii!

FlashAttention (Metal Port)

Everything we actually know about the Apple Neural Engine (ANE)

Apple GPU microarchitecture

LLM101n: Let's build a Storyteller

Efficient Triton Kernels for LLM Training

LLM training in simple, raw C/Metal Shading Language

🚀 Efficient implementations of state-of-the-art linear attention models

Hackable and optimized Transformers building blocks, supporting a composable construction.

Dolphin for iOS, reborn

A port of https://www.github.com/n64decomp/sm64 for modern devices.

ONNX Serving is a project written with C++ to serve onnx-mlir compiled models with GRPC and other protocols.Benefiting from C++ implementation, ONNX Serving has very low latency overhead and high t…

A Super Mario 64 decompilation, brought to you by a bunch of clever folks.

Mirage Persistent Kernel: Compiling LLMs into a MegaKernel

seqax = sequence modeling + JAX

Official implementation of Half-Quadratic Quantization (HQQ)

A CocoaPods plugin to add SPM dependencies to CocoaPods-based projects

DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.