Fork-Merge Decoding: Enhancing Multimodal Understanding in Audio-Visual Large Language Models

TL;DR

This paper introduces Fork-Merge Decoding, an inference-time strategy that improves multimodal understanding in audio-visual large language models by reducing modality bias without additional training.

Contribution

It proposes a simple inference-time method that separates and then merges modality-specific reasoning to enhance balanced multimodal understanding in AV-LLMs.

Findings

Consistent performance improvements across multiple AV-LLMs.

Effective reduction of modality bias during inference.

No additional training or architectural changes needed.

Abstract

The goal of this work is to enhance balanced multimodal understanding in audio-visual large language models (AV-LLMs) by addressing modality bias without additional training. In current AV-LLMs, audio and video features are typically processed jointly in the decoder. While this strategy facilitates unified multimodal understanding, it may introduce modality bias, where the model tends to over-rely on one modality due to imbalanced training signals. To mitigate this, we propose Fork-Merge Decoding (FMD), a simple yet effective inference-time strategy that requires no additional training or architectural modifications. FMD first performs modality-specific reasoning by processing audio-only and video-only inputs through the early decoder layers (fork), and then merges the resulting hidden states for joint reasoning in the remaining layers (merge). This separation allows each modality to be emphasized in the early stages while encouraging balanced contributions during integration. We validate our method on three representative AV-LLMs-VideoLLaMA2, video-SALMONN, and Qwen2.5-Omni-using three benchmark datasets. Experimental results show consistent gains in audio, video, and audio-visual reasoning tasks, highlighting the effectiveness of inference-time interventions for robust and efficient multimodal understanding.