Vibecoding an inference chip
by ankitvgupta on 11/20/2026, 6:15:16 PM
https://ankitg.me/blog/vibecoding-inference-chip/
Comments
by: cpldcpu
I did something like this with Opus 4.5: <a href="https://github.com/cpldcpu/smollm.c" rel="nofollow">https://github.com/cpldcpu/smollm.c</a><p>It seems implementing transformers is the new "hello world" of AI assisted coding.
11/20/2026, 7:51:20 PM
by: chalkycrimp
Good chance I'm not the only one thinking this: share the code
11/20/2026, 7:42:28 PM
by: codekansas
This is cool<p>Next step is tuning your post training pipeline to optimize your model for your chosen FPGA architecture...
11/20/2026, 7:30:03 PM
by: datadrivenangel
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.
11/20/2026, 7:07:57 PM
by: saejox
i wonder if we can emulate an FPGA on a rtx 3090? then test it on an AWS EC2 F2 few hours a day.
11/20/2026, 7:51:31 PM