On Speculative Decoding for Multimodal Large Language Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Gagrani, Mukul, Goel, Raghavv, Jeon, Wonseok, Park, Junyoung, Lee, Mingu, Lott, Christopher
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914753502248960
author Gagrani, Mukul
Goel, Raghavv
Jeon, Wonseok
Park, Junyoung
Lee, Mingu
Lott, Christopher
author_facet Gagrani, Mukul
Goel, Raghavv
Jeon, Wonseok
Park, Junyoung
Lee, Mingu
Lott, Christopher
contents Inference with Multimodal Large Language Models (MLLMs) is slow due to their large-language-model backbone which suffers from memory bandwidth bottleneck and generates tokens auto-regressively. In this paper, we explore the application of speculative decoding to enhance the inference efficiency of MLLMs, specifically the LLaVA 7B model. We show that a language-only model can serve as a good draft model for speculative decoding with LLaVA 7B, bypassing the need for image tokens and their associated processing components from the draft model. Our experiments across three different tasks show that speculative decoding can achieve a memory-bound speedup of up to 2.37$\times$ using a 115M parameter language model that we trained from scratch. Additionally, we introduce a compact LLaVA draft model incorporating an image adapter, which shows marginal performance gains in image captioning while maintaining comparable results in other tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2404_08856
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On Speculative Decoding for Multimodal Large Language Models
Gagrani, Mukul
Goel, Raghavv
Jeon, Wonseok
Park, Junyoung
Lee, Mingu
Lott, Christopher
Computation and Language
Artificial Intelligence
Machine Learning
Inference with Multimodal Large Language Models (MLLMs) is slow due to their large-language-model backbone which suffers from memory bandwidth bottleneck and generates tokens auto-regressively. In this paper, we explore the application of speculative decoding to enhance the inference efficiency of MLLMs, specifically the LLaVA 7B model. We show that a language-only model can serve as a good draft model for speculative decoding with LLaVA 7B, bypassing the need for image tokens and their associated processing components from the draft model. Our experiments across three different tasks show that speculative decoding can achieve a memory-bound speedup of up to 2.37$\times$ using a 115M parameter language model that we trained from scratch. Additionally, we introduce a compact LLaVA draft model incorporating an image adapter, which shows marginal performance gains in image captioning while maintaining comparable results in other tasks.
title On Speculative Decoding for Multimodal Large Language Models
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2404.08856