MLLM-as-a-Judge Exhibits Model Preference Bias

TL;DR

This paper investigates biases in multimodal large language models used as evaluators, revealing tendencies toward self-preference and mutual bias, and proposes a simple ensemble method to mitigate these biases.

Contribution

The study introduces Philautia-Eval to measure model bias, analyzes bias patterns across 12 MLLMs, and proposes Pomms, an ensemble approach that reduces bias while preserving performance.

Findings

Representative MLLMs tend to favor their own outputs.

Mutual preference bias exists within certain model families.

Pomms ensemble reduces bias without sacrificing performance.

Abstract

Automatic evaluation using multimodal large language models (MLLMs), commonly referred to as MLLM-as-a-Judge, has been widely used to measure model performance. If such MLLM-as-a-Judge methods were biased, they could distort model comparisons and benchmark-driven scientific progress. However, it remains unclear to what extent MLLM-as-a-Judge methods favor or disfavor text generated by specific MLLMs. In this study, we propose Philautia-Eval to investigate such model-specific preference bias. Philautia-Eval quantifies the degree of the bias by disentangling preference tendencies from differences in generation quality. Using 1.29M caption-score pairs collected from 12 MLLMs, we found that representative MLLMs tend to exhibit self-preference bias. Moreover, experimental results indicate mutual preference bias within particular model families, which is potentially driven by reused connectors and overlapping instruction-tuning resources. Finally, we introduce a simple ensemble of MLLMs, Pomms. Our results demonstrated that Pomms effectively mitigated the model-specific preference bias while maintaining performance.