Latest

Showcase of a simple way of running custom CUDA kernels in PyTorch with extensions and a quick way of benchmarking them with respect to native functions

A brief overview of some of the techniques that can make a model faster without much code change

Conv 1D is a simple kernel to write, however if you want to optimize it, you will learn about all layers of GPU memory hierarchy.

Implementation of Qwen3 using PyTorch with some inference based optimizations

Beautiful graphics that explain how kernels are executed on the GPU