NHacker Next
  • new
  • past
  • show
  • ask
  • show
  • jobs
  • submit
▲Vibecoding an inference chip (ankitg.me)
cpldcpu 29 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.

codekansas 50 minutes ago [-]
This is cool

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

chalkycrimp 38 minutes ago [-]
Good chance I'm not the only one thinking this: share the code
datadrivenangel 1 hours 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 56 minutes ago [-]
Fpga is software-driven, model can verify its output easily.

Zero surprise it did it

datadrivenangel 1 hours ago [-]
it only cost them $10,000 in Claude API usage. :)
hmate9 1 hours ago [-]
You get that much usage in a monthly $200 subscription.
dvh 29 minutes ago [-]
So who's paying the cost?
exe34 27 minutes ago [-]
Corporate customers
dofm 46 minutes ago [-]
And a kitten.
saejox 29 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.
Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact
Rendered at 20:20:09 GMT+0000 (Coordinated Universal Time) with Vercel.