Blog
Notes on inference, machine learning, and experiments along the way.
Tailored Inference: Why One Engine Won't Be Fastest for Everything
Why fitting inference software to a model, hardware, and workload can outperform a general-purpose serving engine.
Find a Niche by Intersecting Your Strengths
A practical way to find your niche by combining skills you already have.
The DeepSeek V3.2 Breakthrough Simplified
An accessible walkthrough of DeepSeek Sparse Attention and how it reduces attention computation.
A Trick for Backpropagation of Linear Transformations
How to derive gradients for linear transformations expressed with Einstein summation.
How Rubik's Cube Blindfolded Works
A visual introduction to blindfolded solving with the Old Pochmann method.
Rendering Fractals by Optimizing Pixels in JAX
Rendering fractals with JAX, parallel computation, and pixel optimization—with a CUDA follow-up.
Train a Text Summarizer in 5 Minutes
Fine-tune Flan-T5 on XSum with adapter-transformers to build a text summarizer.
Nextcloud Bug Bounty Report (CVE-2020-8155)
The discovery and disclosure of a Nextcloud cross-site scripting vulnerability.