What I’ve written

Long-form notes, every number measured on real hardware.

The projects page shows what I built. This page is for the things I had to work out along the way: how inference actually spends its time, which knobs move the needle and which only look like they do. Each post ships with the scripts and raw results so you can rerun it on your own machine. Hit / to search.

/ 1 post

Featured

llama.cpp Quantization CUDA Python

Getting the most tokens per second out of one laptop with llama.cpp

One Q8_0 model, one laptop, and every knob llama.cpp exposes: thread pinning on hybrid cores, the CPU backend DLLs, per-tensor quantization of the tied LM head, GPU offload splits, KV cache types and self-speculative decoding. Each step measured for speed and for accuracy against the BF16 reference.

September 2026 Long read Read the post

Say hello

Want one of these, but for your problem?

Happy to talk through architecture, trade-offs, or why the obvious approach is not working. Email is fastest.

Email LinkedIn GitHub Twitter Resume