...and the issue was using batch sizes of 64 instead of 16. Now loss looks reasonable.

This is a fantastic tutorial - but when I execute it exactly I run into massive exploding loss…
1
Lee Drake
Lee Drake
·Follow
Jan 24, 2023
--
...and the issue was using batch sizes of 64 instead of 16. Now loss looks reasonable.
--
--
Written by Lee Drake230 Followers
·442 Following
Μη κατατριψης το υπολειπομενον του βιου μερος εν ταις περι ετερων φαντασιαις... ορθον ουν ειναι χρη, ουχι ορθουμενον - Marcus Aurelius
No responses yet
Help
Status
About
Careers
Press
Blog
Privacy
Terms
Text to speech
Teams