Vibecoding an Inference Chip

(ankitg.me)

23 points | by ankitvgupta an hour ago ago

4 comments

  • datadrivenangel 20 minutes ago

    Does not say what type of model or size they were serving, but one of the traces mentions Maple, which is a 20b model. Still, getting ~70+ tok/s is quite good.

    • risyachka 5 minutes ago

      Fpga is software-driven, model can verify its output easily.

      Zero surprise it did it

    • datadrivenangel 20 minutes ago

      it only cost them $10,000 in Claude API usage. :)

      • hmate9 16 minutes ago

        You get that much usage in a monthly $200 subscription.