English

Spec-LLaVA: Accelerating Vision-Language Models with Dynamic Tree-Based Speculative Decoding

Computation and Language 2025-09-16 v1

Abstract

Vision-Language Models (VLMs) enable powerful multimodal reasoning but suffer from slow autoregressive inference, limiting their deployment in real-time applications. We introduce Spec-LLaVA, a system that applies speculative decoding to accelerate VLMs without sacrificing output quality. Spec-LLaVA pairs a lightweight draft VLM with a large target model: the draft speculates future tokens, which the target verifies in parallel, allowing multiple tokens to be generated per step. To maximize efficiency, we design a dynamic tree-based verification algorithm that adaptively expands and prunes speculative branches using draft model confidence. On MS COCO out-of-domain images, Spec-LLaVA achieves up to 3.28×\times faster decoding on LLaVA-1.5 (7B, 13B) with no loss in generation quality. This work presents a lossless acceleration framework for VLMs using dynamic tree-structured speculative decoding, opening a path toward practical real-time multimodal assistants. Importantly, the lightweight draft model design makes the framework amenable to resource-constrained or on-device deployment settings.

Keywords

Cite

@article{arxiv.2509.11961,
  title  = {Spec-LLaVA: Accelerating Vision-Language Models with Dynamic Tree-Based Speculative Decoding},
  author = {Mingxiao Huo and Jiayi Zhang and Hewei Wang and Jinfeng Xu and Zheyu Chen and Huilin Tai and Yijun Chen},
  journal= {arXiv preprint arXiv:2509.11961},
  year   = {2025}
}

Comments

7pages, accepted by ICML TTODLer-FM workshop

R2 v1 2026-07-01T05:36:57.144Z