Efficient and Asymptotically Unbiased Constrained Decoding for Large Language Models

Haotian Ye, Himanshu Jain, Chong You, Ananda Theertha Suresh, Haowei Lin, James Zou, Felix Yu

TL;DR: This paper introduces a novel, efficient constrained decoding algorithm for LLMs that overcomes the inefficiencies and biases of traditional methods.

Abstract: In real-world applications of large language models, outputs are often required to be confined: selecting items from predefined product or document sets, generating phrases that comply with safety standards, or conforming to specialized formatting styles. To control the generation, constrained decoding has been widely adopted. However, existing prefix-tree-based constrained decoding is inefficient under GPU-based model inference paradigms, and it introduces unintended biases into the output distribution. This paper introduces Dynamic Importance Sampling for Constrained Decoding (DISC) with GPU-based Parallel Prefix-Verification (PPV), a novel algorithm that leverages dynamic importance sampling to achieve theoretically guaranteed asymptotic unbiasedness and overcomes the inefficiency of prefix-tree. Extensive experiments demonstrate the superiority of our method over existing methods in both efficiency and output quality. These results highlight the potential of our methods to improve constrained generation in applications where adherence to specific constraints is essential.

Submission Number: 1709