VidLaDA: Bidirectional Diffusion Large Language Models for Efficient Video Understanding
Zhihao He, Tieyuan Chen, Kangyu Wang, Ziran Qin, Yang Shao, Chaofan Gan, Shijie Li, Zuxuan Wu
Abstract
Current Video Large Language Models (Video LLMs) typically encode frames via a vision encoder and employ an autoregressive (AR) LLM for understanding and generation. However, this AR paradigm inevitably faces a dual efficiency bottleneck: strictly unidirectional attention compromises *understanding efficiency* by hindering global spatiotemporal aggregation, while serial decoding restricts *generation efficiency*. To address this, we propose **VidLaDA**, a Video LLM based on Diffusion Language Models (DLMs) that leverages bidirectional attention to unlock comprehensive spatiotemporal modeling and decode tokens in parallel. To further mitigate the computational overhead of diffusion decoding, we introduce **MARS-Cache**, an acceleration strategy that prunes redundancy by combining asynchronous visual cache refreshing with frame-wise chunk attention. Experiments show VidLaDA rivals state-of-the-art AR baselines (e.g., Qwen2.5-VL and LLaVA-Video) and outperforms DLM baselines, with MARS-Cache delivering over 12x speedup without compromising accuracy. *Code and checkpoints will be available in the camera-ready version.*
BibTeX
@inproceedings{
he2026vidlada,
title={VidLa{DA}: Bidirectional Diffusion Large Language Models for Efficient Video Understanding},
author={Zhihao He and Tieyuan Chen and Kangyu Wang and Ziran Qin and Yang Shao and Chaofan Gan and Shijie Li and Zuxuan Wu and Weiyao Lin},
booktitle={Forty-third International Conference on Machine Learning},
year={2026},
url={https://openreview.net/forum?id=efx3dOgAYw}
}