Paper ID: 2406.12295

Fast and Slow Generating: An Empirical Study on Large and Small Language Models Collaborative Decoding

Kaiyan Zhang, Jianyu Wang, Ning Ding, Biqing Qi, Ermo Hua, Xingtai Lv, Bowen Zhou

Large Language Models (LLMs) exhibit impressive capabilities across various applications but encounter substantial challenges such as high inference latency, considerable training costs, and the generation of hallucinations. Collaborative decoding between large and small language models (SLMs) presents a promising strategy to mitigate these issues through methods including speculative decoding, contrastive decoding, and emulator or proxy fine-tuning. However, the specifics of such collaborations, particularly from a unified perspective, remain largely unexplored. Inspired by dual-process cognitive theory, we propose a unified framework in this paper, termed Fast and Slow Generating (FS-GEN). Within this framework, LLMs (sometimes along with SLMs) are categorized as System 2 (slow and deliberate), while independent SLMs are designated as System 1 (fast and intuitive). We provide a comprehensive analysis of these collaborative methodologies, elucidating their common properties and shedding light on the differential knowledge capabilities of System 2 versus System 1 through the FS-GEN framework. Our findings indicate that only a small proportion of collaborative interactions (approximately less than 20\% in most instances) are necessary across various methods. These interactions between System 1 and System 2 conform to a scaling law related to the parameter ratios, enabling predictable collaboration. Furthermore, we explore the specific conditions under which collaboration proves most effective, particularly from an uncertainty perspective, offering novel insights that may guide future optimization efforts. Our research underscores that the fundamental distinction between System 1 and System 2 lies in the uncertainty of next token predictions, where interventions by System 2 are crucial to support System 1. Code for Reproduction: this https URL

Submitted: Jun 18, 2024