MASSV: Multimodal Adaptation and Self-Data Distillation for Speculative Decoding of Vision-Language Models

TL;DR

MASSV introduces a scalable method to accelerate vision-language models by transforming small language models into effective multimodal drafters through adaptation and self-distillation, significantly improving inference speed and output length.

Contribution

It presents a novel two-phase approach to adapt small language models for multimodal speculative decoding in VLMs, addressing key architectural and prediction alignment challenges.

Findings

Increases accepted length by up to 30%.

Achieves up to 1.46x inference speedup.

Effective across multiple VLM model families.

Abstract

Speculative decoding significantly accelerates language model inference by enabling a lightweight draft model to propose multiple tokens that a larger target model verifies simultaneously. However, applying this technique to vision-language models (VLMs) presents two fundamental challenges: small language models that could serve as efficient drafters lack the architectural components to process visual inputs, and their token predictions fail to match those of VLM target models that consider visual context. We introduce Multimodal Adaptation and Self-Data Distillation for Speculative Decoding of Vision-Language Models (MASSV), which transforms existing small language models into effective multimodal drafters through a two-phase approach. MASSV first connects the target VLM's vision encoder to the draft model via a lightweight trainable projector, then applies self-distilled visual instruction tuning using responses generated by the target VLM to align token predictions. Comprehensive experiments across the Qwen2.5-VL and Gemma3 model families demonstrate that MASSV increases accepted length by up to 30% and delivers end-to-end inference speedups of up to 1.46x on visually-grounded tasks. MASSV provides a scalable, architecture-compatible method for accelerating both current and future VLMs.