SmolVLM: Redefining small and efficient multimodal models

TL;DR

SmolVLM introduces a series of resource-efficient multimodal models that outperform larger models in image and video tasks while using significantly less GPU memory, enabling practical deployment on edge devices.

Contribution

The paper presents novel architectural and tokenization strategies for small VLMs, achieving high performance with minimal memory usage, and demonstrates their effectiveness on image and video tasks.

Findings

SmolVLM-256M uses less than 1GB GPU memory during inference.

SmolVLM outperforms much larger models like Idefics-80B despite smaller size.

Models show strong video comprehension capabilities.

Abstract

Large Vision-Language Models (VLMs) deliver exceptional performance but require significant computational resources, limiting their deployment on mobile and edge devices. Smaller VLMs typically mirror design choices of larger models, such as extensive image tokenization, leading to inefficient GPU memory usage and constrained practicality for on-device applications. We introduce SmolVLM, a series of compact multimodal models specifically engineered for resource-efficient inference. We systematically explore architectural configurations, tokenization strategies, and data curation optimized for low computational overhead. Through this, we identify key design choices that yield substantial performance gains on image and video tasks with minimal memory footprints. Our smallest model, SmolVLM-256M, uses less than 1GB GPU memory during inference and outperforms the 300-times larger Idefics-80B model, despite an 18-month development gap. Our largest model, at 2.2B parameters, rivals state-of-the-art VLMs consuming twice the GPU memory. SmolVLM models extend beyond static images, demonstrating robust video comprehension capabilities. Our results emphasize that strategic architectural optimizations, aggressive yet efficient tokenization, and carefully curated training data significantly enhance multimodal performance, facilitating practical, energy-efficient deployments at significantly smaller scales.