Search papers, labs, and topics across Lattice.
This paper introduces OPD-V, a visual On-Policy Self-Distillation method that addresses the challenge of Modality Imbalance in multimodal large language models (MLLMs). By leveraging Positive and Negative Teachers to create a Modality-Balance Trust Region, the approach enhances the integration of multimodal inputs during self-distillation. Experimental results across multiple benchmarks demonstrate that OPD-V significantly improves reasoning performance while also reducing training costs compared to existing methods.
Modality Balance can be harnessed as a powerful form of privileged information, leading to substantial gains in reasoning performance for multimodal models.
On-Policy Self-Distillation (OPSD) has become a standard post-training approach for improving visual reasoning in multimodal large language models (MLLMs). Existing methods draw privileged information from diverse input sources to guide self-distillation. Yet these designs overlook Modality Imbalance, a challenge inherent to MLLM reasoning. When textual information dominates generation, the model cannot fully integrate its multimodal input. Consequently, carefully designed privileged information remains underused, limiting the effectiveness of OPSD. To examine this limitation, we construct a Positive Teacher with the Zoom-In Image and a Negative Teacher with the Mask Image, which exhibit different degrees of Modality Imbalance. Changes in their reasoning correctness and token logits reveal that Modality Balance can itself serve as privileged information. Motivated by this finding, we introduce OPD-V, a visual OPSD paradigm that instantiates such information through the Positive Teacher and Negative Teacher. Positive Modality-Balance Logits Margins define a Modality-Balance Trust Region that selects the on-policy tokens used for self-distillation. Experiments across 6 benchmarks, 4 MLLM backbones, and 5 post-training methods show that OPD-V consistently improves reasoning performance while reducing training cost.
NUS