Some things are more CRINGE than others: Iterative Preference Optimization with the Pairwise Cringe Loss

arXiv:2312.16682v2 Announce Type: replace
Abstract: Practitioners commonly align large language models using pairwise preferences, i.e., given labels of the type response A is preferred to response B for a given input. Perhaps less commonly, methods have also been developed for binary feedback, i.e. training models given labels of type response A is good or bad. We show how an existing performant binary feedback method, the Cringe Loss (Adolphs et al., 2022), can be generalized to the pairwise preference setting using …

abstract arxiv binary cs.ai cs.cl feedback iterative labels language language models large language large language models loss optimization training training models type