Vibecoding an inference chip (ankitg.me)

51 points by ankitvgupta 2 hours ago

cpldcpu 37 minutes ago

I did something like this with Opus 4.5: https://github.com/cpldcpu/smollm.c

It seems implementing transformers is the new "hello world" of AI assisted coding.

datadrivenangel an hour 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 an hour ago

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

Zero surprise it did it

datadrivenangel an hour ago

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

hmate9 an hour ago

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

dvh 37 minutes ago

dofm an hour ago

codekansas an hour ago

This is cool

Next step is tuning your post training pipeline to optimize your model for your chosen FPGA architecture...

chalkycrimp an hour ago

Good chance I'm not the only one thinking this: share the code

Archit3ch 4 minutes ago

Now do silicon. ;)

saejox 37 minutes ago

i wonder if we can emulate an FPGA on a rtx 3090? then test it on an AWS EC2 F2 few hours a day.