RLHF improves OOD generalization over SFT especially under larger distribution shifts but reduces output diversity, revealing a tradeoff in LLM fine-tuning methods.
Either an imperative sentence or a question is permitted
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2023 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Understanding the Effects of RLHF on LLM Generalisation and Diversity
RLHF improves OOD generalization over SFT especially under larger distribution shifts but reduces output diversity, revealing a tradeoff in LLM fine-tuning methods.