Hacker News new | ask | show | jobs
by tysam_and 1231 days ago
"hackable" and "simple yet complete technical introduction"

Music to my ears, well done and don't worry too much about the negative comments! They'll come out for anything you do I think.

I saw a tweet from someone the other day talking about how they massively increased their training speed by changing part of their architecture to have dimensions that were a factor of 64 rather than a prime-like kind of number.

One of the comments below it? ~"Seems very architecture specific."

lol.

So don't sweat it! <3 Great work and thanks for putting yourself out there, super job! :D :D :D :D :)))))) <3 :D :D :fireworks:

1 comments

We do GPU-specific training and inference speedups, at CentML.
Grata, well deserved.