Temperature-Centric Investigation of Speculative Decoding with Knowledge Distillation
This work addresses a specific bottleneck in speculative decoding for language model inference, offering incremental improvements in understanding and optimizing generation configurations.
The paper investigates how decoding temperatures affect the efficiency of speculative decoding in language models, finding that knowledge distillation in consistent temperature settings can mitigate challenges at higher temperatures and improve speedup.
Speculative decoding stands as a pivotal technique to expedite inference in autoregressive (large) language models. This method employs a smaller draft model to speculate a block of tokens, which the target model then evaluates for acceptance. Despite a wealth of studies aimed at increasing the efficiency of speculative decoding, the influence of generation configurations on the decoding process remains poorly understood, especially concerning decoding temperatures. This paper delves into the effects of decoding temperatures on speculative decoding's efficacy. Beginning with knowledge distillation (KD), we first highlight the challenge of decoding at higher temperatures, and demonstrate KD in a consistent temperature setting could be a remedy. We also investigate the effects of out-of-domain testing sets with out-of-range temperatures. Building upon these findings, we take an initial step to further the speedup for speculative decoding, particularly in a high-temperature generation setting. Our work offers new insights into how generation configurations drastically affect the performance of speculative decoding, and underscores the need for developing methods that focus on diverse decoding configurations. Code is publically available at https://github.com/ozyyshr/TempSpec.