The Impact of Inference Acceleration on Bias of LLMs

TL;DR

This paper investigates how inference acceleration techniques like quantization and pruning affect demographic bias in Large Language Models, revealing complex and unpredictable bias changes that require careful evaluation.

Contribution

It is the first comprehensive analysis of how inference acceleration strategies impact bias in LLMs, emphasizing the need for bias assessment post-optimization.

Findings

Bias changes significantly after acceleration methods

Bias effects are complex and model-dependent

Evaluation of bias must be case-by-case

Abstract

Last few years have seen unprecedented advances in capabilities of Large Language Models (LLMs). These advancements promise to benefit a vast array of application domains. However, due to their immense size, performing inference with LLMs is both costly and slow. Consequently, a plethora of recent work has proposed strategies to enhance inference efficiency, e.g., quantization, pruning, and caching. These acceleration strategies reduce the inference cost and latency, often by several factors, while maintaining much of the predictive performance measured via common benchmarks. In this work, we explore another critical aspect of LLM performance: demographic bias in model generations due to inference acceleration optimizations. Using a wide range of metrics, we probe bias in model outputs from a number of angles. Analysis of outputs before and after inference acceleration shows significant change in bias. Worryingly, these bias effects are complex and unpredictable. A combination of an acceleration strategy and bias type may show little bias change in one model but may lead to a large effect in another. Our results highlight a need for in-depth and case-by-case evaluation of model bias after it has been modified to accelerate inference.