OVM, Outcome-supervised Value Models for Planning in Mathematical Reasoning

arXiv:2311.09724v2 Announce Type: replace-cross
Abstract: Large language models (LLMs) often struggle with maintaining accuracy throughout multiple multiple reasoning steps, especially in mathematical reasoning where an error in earlier steps can propagate to subsequent ones and it ultimately leading to an incorrect answer. To reduce error propagation, guided decoding is employed to direct the LM decoding on a step-by-step basis. We argue that in guided decoding, assessing the potential of an incomplete reasoning path can be more advantageous than simply ensuring …

abstract accuracy arxiv cs.ai cs.cl decoding error language language models large language large language models llms mathematical reasoning multiple planning propagation reasoning reduce struggle type value