Luspo Length Unbiased Sequences

Fix length bias in sequence-level policy optimization by scaling each sequence's loss by its token count, eliminating gradient imbalances that cause models to shorten responses during training.

adu2021 Updated 7 repo stars

File contents

adu2021/skillxiv commit 66952bb90d

Frequently asked questions

npx get-skill add adu2021/luspo-length-unbiased-sequences