6 ms·

etched tried this.... it didn't go very well

• anukin 22d ago
  I would assume asic based llm would work really well. Why did it not go well?
  • striking 22d ago
    https://chatjimmy.ai/ https://chatjimmy.ai/ runs Llama 3.1-8B on an ASIC as a demo by https://taalas.com/ https://taalas.com/ I believe. That's quite a few parameters shy of today's trillion-weight behemoths, but it is fast.
    • mchusma 22d ago
      You are correct. I think this is the bull case. It seems like this would be useful right now for some things (eg moderation).