Comparing Bad Apples to Good Oranges Aligning Large Language Models via Joint Preference Optimization

A common technique for aligning large language models (LLMs) relies on acquiring human preferences by comparing multiple generations conditioned on a fixed context.This method, however, relies solely on pairwise comparisons, where the generations are evaluated within an identical context.While effective to such conditional preferences often fail to encompass the nuanced and multidimensional nature of human preferences.In this work, we revisit the traditional paradigm of preference acquisition and propose a new axis based on eliciting preferences jointly over the instructionresponse pairs.Unlike prior preference optimizations, which are designed for conditional ranking protocols (e.g., DPO), we propose Joint Preference Optimization (JPO), a new preference optimization objective that upweights the joint probability of the chosen instructionresponse pair over the rejected instructionresponse pair.Interestingly, LLMs trained with joint instruction-response preference data using JPO outperform LLM trained with DPO by 5.2% and 3.3% win-rate for summarization and open-ended dialogue datasets, respectively.Our findings reveal that joint preferences over instruction and response pairs can significantly enhance the alignment of LLMs by tapping into a broader spectrum of human preference elicitation.The data and code is available at https://github.com/Hritikbansal/jpo.Create a list of four fruits other than Apple Create a list of four fruits Apple, Orange, Banana, Grape Orange, Blueberry, Kiwi, Banana Apple, Orange, Banana, Grape Create a list of beach activities wear sunscreen, don't litter