DPO with backtranslation post-training raises English-to-German COMET from 0.703 to 0.747 on gemma3-1b.
ORPO: Monolithic Preference Optimization without Reference Model
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Backtranslation Augmented Direct Preference Optimization for Neural Machine Translation
DPO with backtranslation post-training raises English-to-German COMET from 0.703 to 0.747 on gemma3-1b.