A GPU defined in software. Runs Llama 3.2 1B at 3.6 tok/sec. Zero dependencies. - View it on GitHub