transformers
Fix how we compute the final non-padding token for ForSequenceClassification models
#35911
Merged
Merged
Fix how we compute the final non-padding token for ForSequenceClassification models #35911
Rocketknight1
merged
12 commits into
main
from
fix_sequence_classification_padding_side
Fix how we compute the final non-padding token for Gemma (and probabl…
f74d4dda
.size() -> .shape[]
067d99a2
Propagating changes to other models
3e381f41
Propagating changes to other models
7cc13967
Change it for all ForSequenceClassification models
d3b7c994
Fix batch dim
1c11edc4
More TF fixes
13a670ee
Copy the TF fix around as well
c671810f
Correct layer name for TFCTRL
8ccde63d
Cleaner .to()
8c69579f
Clean up the nested if-else
172cfd76
Use argmax() instead of .max().values
26d554e3
Cyrilvallez
approved these changes
on 2025-01-30
Login to write a write a comment.
Login via GitHub