I see what you're saying, and it all sounds correct. Doing it as a concatenation seemed easier with the way my code is set up, but when I have time and if my net structure is still similar, I will try that out as optimization.connor_mcmonigle wrote: ↑Fri Aug 27, 2021 10:48 pm Hopefully my description is clear. This should be equivalent, enable for more KPM cache entries (improve the hit rate) and also be somewhat faster given a KPM cache hit as you only have to perform a 480x16 matrix vector product.
Some more experiments with neural nets
Moderators: hgm, Rebel, chrisw
-
- Posts: 178
- Joined: Wed Nov 13, 2019 1:36 am
- Full name: Jonathan Kreuzer