hi, i wrote some trivial garbage that improves llm inference bandwidth by huge%, because this entire trillion dollar industry is idiots. today i'm doing more benchmarks to put on graphs to put in front of the idiots