NumKong: 2'000 Mixed Precision Kernels for All

#1

Post by ashvardanian » Fri, Mar 20, 2026, 7:29 PM UTC

NumKong: 2'000 Mixed Precision Kernels for All
ashvardanian.com

Re: NumKong: 2'000 Mixed Precision Kernels for All

#2

Post by jmalicki » Sat, Mar 21, 2026, 2:32 AM UTC

I wish this had a README.md written by a human so I could understand what the project was about, since it sounds cool on the surface.

Re: NumKong: 2'000 Mixed Precision Kernels for All

#3

Post by ashvardanian » Sat, Mar 21, 2026, 5:56 AM UTC

jmalicki wrote: post #2

I wish this had a README.md written by a human so I could understand what the project was about, since it sounds cool on the surface.

The README was written by a human. I’ve used models extensively to refine the content, but never accepted more than a couple of lines of edits at a time.

Re: NumKong: 2'000 Mixed Precision Kernels for All

#4

Post by jmalicki » Sat, Mar 21, 2026, 6:00 AM UTC

ashvardanian wrote: post #3

jmalicki wrote: post #2

I wish this had a README.md written by a human so I could understand what the project was about, since it sounds cool on the surface.

The README was written by a human. I’ve used models extensively to refine the content, but never accepted more than a couple of lines of edits at a time.

I want to love this project - but can you edit that to give a more succinct summary of what it is without the hype building?

Re: NumKong: 2'000 Mixed Precision Kernels for All

#5

Post by Archit3ch » Sat, Mar 21, 2026, 11:02 AM UTC

Is anyone doing audio processing on Apple SME?

Re: NumKong: 2'000 Mixed Precision Kernels for All

#6

Post by ashvardanian » Sat, Mar 21, 2026, 5:51 PM UTC

Archit3ch wrote: post #5

Is anyone doing audio processing on Apple SME?

I'm not aware of that, but it would likely be a great application area for SME!

Re: NumKong: 2'000 Mixed Precision Kernels for All

#7

Post by saberience » Sun, Mar 22, 2026, 1:25 PM UTC

This is one of those things that sounds mind-bogglingly complex and no doubt impressive... but the article is so long (I tried to read it all) and so complex, that I feel only about 5 people in the world probably can actually understand it, or when it can actually be useful.

You're clearly a brilliant engineer, but I would suggest it would be good if you could write something much, much shorter, easier to understand for those (majority of engineers) who didn't study low level chip design. And also have a handful of clear examples of when you would use this...

So it's like nice job man... but this is so complicated and hard to understand I have no idea how to use it, when I would use it, etc etc. Like I feel I would need to study for several years to grok this article, and well, I don't have time of that sadly.