Blog Archive

You Could Have Come Up with Speculative Decoding★

Building up speculative decoding from first principles

The Boon of Dimensionality

or: Why High Dimensions work for ML

Twelve Attempts at an FP4 Kernel

A worklog of NVFP4 kernels, failed experiments, and one stubborn memory bus

An introduction to CuTe's Layout Algebra★

Hierarchical Navigable Small Worlds★

A deep dive into HNSW, an approximate nearest neighbor search algorithm.

The Rabbit Hole of Building a Filesystem Watcher★

Intuition behind Power of 2 Choices Load balancing

Exploring Clojure for Advent of Code

Making Machine Learning models smaller★