Benchmarking the Energy Savings with Speculative Decoding Strategies

TL;DR

This paper surveys the energy consumption of speculative decoding strategies in large language models, analyzing how factors like model size, decoding methods, and datasets impact energy efficiency.

Contribution

It provides a comprehensive analysis of energy requirements in speculative decoding, highlighting key factors influencing energy savings in LLM inference.

Findings

Energy consumption varies significantly with model size and decoding strategy.

Certain speculative decoding strategies offer notable energy savings.

Dataset characteristics influence the effectiveness of energy optimization.

Abstract

Speculative decoding has emerged as an effective method to reduce latency and inference cost of LLM inferences. However, there has been inadequate attention towards the energy requirements of these models. To address this gap, this paper presents a comprehensive survey of energy requirements of speculative decoding strategies, with detailed analysis on how various factors -- model size and family, speculative decoding strategies, and dataset characteristics -- influence the energy optimizations.