Show HN: MicroLLM lab – try small LMs in browser, see tok/s and accuracy (github.com)

I saw some great results from really tiny LLM's, and ended up making a little benchmark so I could test them in the browser. Chrome, Firefox, Safari and Edge are all supported via web-GPU. I got very fast speeds 30-40 tok/sec even on limited hardware. Let me know what you think!