Quasi-random Multi-Sample Inference for Large Language Models

TL;DR

This paper introduces arithmetic sampling, a quasi-random multi-sample inference method for large language models that enhances diversity and accuracy in tasks like reasoning and translation without extra computational cost.

Contribution

It presents a novel arithmetic sampling technique leveraging the implicit code book of LLMs, improving multi-sample inference efficiency and diversity over traditional methods.

Findings

Arithmetic sampling yields more diverse samples.

Significant accuracy improvements on reasoning and translation tasks.

No additional computational overhead compared to traditional methods.

Abstract

Large language models (LLMs) are often equipped with multi-sample decoding strategies. An LLM implicitly defines an arithmetic code book, facilitating efficient and embarrassingly parallelizable \textbf{arithmetic sampling} to produce multiple samples using quasi-random codes. Traditional text generation methods, such as beam search and sampling-based techniques, have notable limitations: they lack parallelizability or diversity of sampled sequences. This study explores the potential of arithmetic sampling, contrasting it with ancestral sampling across two decoding tasks that employ multi-sample inference: chain-of-thought reasoning with self-consistency and machine translation with minimum Bayes risk decoding. Our results demonstrate that arithmetic sampling produces more diverse samples, significantly improving reasoning and translation performance as the sample size increases. We observe a $\mathbf{3\text{-}5\%}$ point increase in accuracy on the GSM8K dataset and a $\mathbf{0.45\text{-}0.89\%}$ point increment in COMET score for WMT19 tasks using arithmetic sampling without any significant computational overhead.