LLM training in simple, raw C/CUDA

Instant neural graphics primitives: lightning fast NeRF and more

A massively parallel, optimal functional runtime in Rust

FSA/FST algorithms, differentiable, with PyTorch compatibility.

Efficient GPU kernels for block-sparse matrix multiplication and convolution

Fast CUDA matrix multiplication from scratch

Distributed multigrid linear solver library on GPU

PopSift is an implementation of the SIFT algorithm in CUDA.

A UNIVERSAL MUSIC TRANSLATION NETWORK - a method for translating music across musical instruments and styles.

State of the art sorting and segmented sorting, including OneSweep. Implemented in CUDA, D3D12, and Unity style compute shaders. Theoretically portable to all wave/warp/subgroup sizes.

CUDA Matrix Factorization Library with Alternating Least Square (ALS)

GPU Recursive Filtering

WIP · CUDA compatibility for Blaze · https://bitbucket.org/blaze-lib/blaze

Teaching materials, slides and exercises, for the GPU & CUDA training in 2017

research project