davenull
abusing.technology
Kubernetes, mesh radio, keyboards, locks and birds. Notes on the things I take apart, and on the ones that win.
Many computers were hurt in this production.
Recent posts
The audio thread cannot wait
birdsynth is a wavetable synth with a Rust engine running in an AudioWorklet. The interesting part is what the real-time constraint does to the design, and how you write a test that proves you obeyed it.
Where a token actually spends its time
A new explainer that follows one token through a GPU cluster, and makes the case that prefill and decode are different problems sharing hardware.
A site you update with git push
The old abusing.technology was a CMS starter I never finished configuring. The new one has no CMS, no login and no server, just markdown in a repo.
Watching GPT-2 pick a word, one step at a time
An explainer that runs a real model in your browser and shows the eight steps between a prompt and a single token, including the one everybody gets wrong.
Bridging APRS and Meshtastic without losing the replies
Relaying packets one way is easy. Getting a reply back to the right device, for the right operator, is where the interesting problems live.
Active projects
birdsynth
A wavetable synthesizer for the browser with the feature set of a modern software synth, plus an explainer layer that shows each stage working on live signals from the engine. The DSP is Rust compiled to WebAssembly (simd128, no imports) running in an AudioWorklet, behind a Svelte 5 UI.
Meshdash APRS ↔ Meshtastic bridge
A Meshdash relay plugin giving full bi-directional comms between APRS and Meshtastic, plus node registration so APRS users can reach the DMs of all their devices.
GPU token explainer
Sixteen stages in five acts, following one token from an HTTPS request hitting a load balancer, through a GPU cluster's scheduler, sharded weights and collectives, into the transformer math and back out as streamed text. Every cluster figure is real arithmetic over the configuration you set.
LLM token explainer
A step-by-step walkthrough of how a language model produces one token: tokenization, embedding, attention, logits, sampling. Runs GPT-2 in the browser, so the numbers on screen are the real ones.
pyMC_Repeater
A lightweight MeshCore repeater daemon in Python, built on pymc_core.
pyMC_core
A Python MeshCore library with SPI LoRa radio support.