RPO reverses the constraints a response fails to turn any sampled response into a perfect chosen example, improving DPO on multi-constraint instruction following by 2.5 to 4.6 points.
at least three profession-related keywords must appear
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Reverse Preference Optimization for Complex Instruction Following
RPO reverses the constraints a response fails to turn any sampled response into a perfect chosen example, improving DPO on multi-constraint instruction following by 2.5 to 4.6 points.