OPD-V is introduced, a visual OPSD paradigm that instantiates privileged information through the Positive Teacher and Negative Teacher that consistently improves reasoning performance while reducing training cost.
Abstract
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.
On-policy self-distillation (OPSD) is promising as it removes the external teacher required by on-policy distillation (OPD), yet it still needs asymmetric information between teacher and student to ensure that the self-teacher provides a stronger learning signal than the student. Existing methods create this asymmetry either through privileged answers or visual evidence. We ask whether both can be removed, yielding a simpler form of OPSD driven purely by input conditioning. For this purpose, we propose Visual Contrastive Self-Distillation, namely VCSD, which converts image-content removal into an on-policy self-distillation signal. At each student-generated response prefix, the EMA teacher produces two next-token distributions under the same prompt and prefix -- one conditioned on the original image and the other on a content-erased control. Their token-wise log-probability difference highlights candidates whose likelihood is specifically increased by the instance-level visual content. We use this contrast to sharpen the teacher's original-image distribution within its plausible support, and distill the resulting full-distribution target into the student. Using ViRL39K dataset, VCSD consistently outperforms matched OPSD across Qwen3-VL and Qwen3.5 models. For example, on Qwen3-VL, it improves the seven-benchmark aggregate from $62.27\% \rightarrow 67.04\%$ at 2B, $71.30\% \rightarrow 73.16\%$ at 4B, and $72.51\% \rightarrow 76.26\%$ at 8B. Furthermore, VCSD requires no external teacher, privileged answers, visual evidence signals, reasoning traces, or additional inference-time cost.
Yijun Liang, Yunjie Tian, Yijiang Li et al.· arXiv.org· 4 citations
It is demonstrated that resolution differences can serve as a simple and scalable source of privileged information, providing an effective and efficient approach to on-policy self-distillation for multimodal large language models.
This work introduces CVPD (Contrastive Counterfactual Visual Process Distillation), which is the first fully self-contained framework for dense, on-policy, token-level visual self-distillation for MLLMs, and proposes a three-gate Counterfactual Criterion that identifies visual blind spots where zooming into a region changes and sharpens the model's answer distribution.
Shravan Venkatraman, Omkar Thawakar, Ritesh Thawkar et al.· 1 citation
On-Policy Omni Distillation (OPOD), which consolidates text, image, and audio teachers into one omni model, and surpasses the base model and pooled RL training on all twelve benchmarks, and ranks first or second on eleven even when the teachers are included.
Tong Zhao, Yuyang Hu, Reed Li et al.· arXiv.org· 0 citations
Self-Supervised Visual On-Policy Distillation (S$^2$VOPD), a simple yet effective method that constructs on-policy learning signals from asymmetric augmented views, systematically explores a broad design space of visual augmentations and uncover that asymmetry matters.
Yijiang Li, Yijun Liang, Yunjie Tian et al.· 0 citations
Experiments across video understanding and reasoning benchmarks show that the Evidence-Grounded Self-Teacher framework consistently improves upon Standard OPSD across multiple backbones and achieves performance comparable to GRPO while requiring substantially less training time, establishing an effective and efficient post-training approach for Video-LLMs.
Zi-Yue Wang, Shiqi Huang, Wei-Wen Xu et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
