Saved in:
Bibliographic Details
Main Authors: Li, Songlin, Zhu, Xin, Guan, Zechao, Chen, Peipeng, Yao, Jian
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.11423
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917334657007616
author Li, Songlin
Zhu, Xin
Guan, Zechao
Chen, Peipeng
Yao, Jian
author_facet Li, Songlin
Zhu, Xin
Guan, Zechao
Chen, Peipeng
Yao, Jian
contents Traditional black-box distillation for Large Vision-Language Models (LVLMs) typically relies on a single teacher response per input, which often yields high-variance responses and format inconsistencies in multimodal or temporal scenarios. To mitigate this unreliable supervision, we propose R-MSD (Reliable Multi-Sample Distillation), a framework that explicitly models teacher sampling variance to enhance distillation stability. Rather than relying on a single teacher response, our approach leverages a task-adaptive teacher pool to provide robust supervision tailored to both closed-ended and open-ended reasoning. By integrating quality-aware signal matching with an adversarial distillation objective, our approach effectively filters teacher noise while maximizing knowledge transfer. Extensive evaluations across comprehensive video understanding benchmarks demonstrate that R-MSD consistently outperforms single sample distillation methods. We additionally include an original SFT+RL 4B baseline under the same training budget, which shows only marginal gains, while our method achieves significant improvements. With a 4B student model, our approach delivers gains on VideoMME (+1.5%), Video-MMMU (+3.2%), and MathVerse (+3.6%).
format Preprint
id arxiv_https___arxiv_org_abs_2603_11423
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Beyond Single-Sample: Reliable Multi-Sample Distillation for Video Understanding
Li, Songlin
Zhu, Xin
Guan, Zechao
Chen, Peipeng
Yao, Jian
Computer Vision and Pattern Recognition
Traditional black-box distillation for Large Vision-Language Models (LVLMs) typically relies on a single teacher response per input, which often yields high-variance responses and format inconsistencies in multimodal or temporal scenarios. To mitigate this unreliable supervision, we propose R-MSD (Reliable Multi-Sample Distillation), a framework that explicitly models teacher sampling variance to enhance distillation stability. Rather than relying on a single teacher response, our approach leverages a task-adaptive teacher pool to provide robust supervision tailored to both closed-ended and open-ended reasoning. By integrating quality-aware signal matching with an adversarial distillation objective, our approach effectively filters teacher noise while maximizing knowledge transfer. Extensive evaluations across comprehensive video understanding benchmarks demonstrate that R-MSD consistently outperforms single sample distillation methods. We additionally include an original SFT+RL 4B baseline under the same training budget, which shows only marginal gains, while our method achieves significant improvements. With a 4B student model, our approach delivers gains on VideoMME (+1.5%), Video-MMMU (+3.2%), and MathVerse (+3.6%).
title Beyond Single-Sample: Reliable Multi-Sample Distillation for Video Understanding
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.11423