0 0 0 Python

Fix: ibotpluplus loss normalisation is (#1937)

* Ensure loss is normalised across masked and non masked patch tokens

* Add additional tests to make sure masking loss normalisation is correct

* Formatting

Igor Susmelj committed 3mo ago

c90f3960ac484ee3602076fc6418a7d23d1acb2a

Parent: a08c5c3

Committed by GitHub <noreply@github.com> on 6/8/2026, 2:34:13 PM