Improving Minimum Bayes Risk Decoding with Multi-Prompt

TL;DR

This paper introduces multi-prompt decoding combined with Minimum Bayes Risk to enhance text generation quality by leveraging diverse prompts, leading to more stable and higher-quality outputs across various tasks and models.

Contribution

The paper proposes multi-prompt decoding with MBR, improving generation stability and quality by capturing diverse candidate outputs, addressing prompt sensitivity issues.

Findings

Multi-prompt decoding improves MBR performance across tasks.

Diverse prompts lead to higher quality candidate sets.

Multi-prompt enhances generation across models and metrics.

Abstract

While instruction fine-tuned LLMs are effective text generators, sensitivity to prompt construction makes performance unstable and sub-optimal in practice. Relying on a single "best" prompt cannot capture all differing approaches to a generation problem. Using this observation, we propose multi-prompt decoding, where many candidate generations are decoded from a prompt bank at inference-time. To ensemble candidates, we use Minimum Bayes Risk (MBR) decoding, which selects a final output using a trained value metric. We show multi-prompt improves MBR across a comprehensive set of conditional generation tasks, and show this is a result of estimating a more diverse and higher quality candidate space than that of a single prompt. Further experiments confirm multi-prompt improves generation across tasks, models and metrics.