Skip to content

feat(trainer): KL penalty over unmasked tokens only#2458

Closed
samsja wants to merge 1 commit into
feat/dppo-diff-default-lossfrom
feat/dppo-kl-on-unmasked
Closed

feat(trainer): KL penalty over unmasked tokens only#2458
samsja wants to merge 1 commit into
feat/dppo-diff-default-lossfrom
feat/dppo-kl-on-unmasked

Commits

Commits on May 9, 2026