UVLM: A Universal Vision-Language Model Loader for Reproducible Multimodal Benchmarking

TL;DR

UVLM is a versatile, open-source framework that unifies the loading, configuration, and benchmarking of diverse vision-language models on custom tasks, promoting reproducibility and comparability in multimodal research.

Contribution

The paper introduces UVLM, a unified, Google Colab-compatible framework supporting multiple VLM architectures with standardized evaluation tools for the first comprehensive benchmarking on complex reasoning tasks.

Findings

Supported models include LLaVA-NeXT and Qwen2.5-VL.

Benchmarking conducted on 120 street-view images with increasing reasoning complexity.

UVLM enables consistent comparison of VLMs using identical prompts and evaluation protocols.

Abstract

Vision-Language Models (VLMs) have emerged as powerful tools for image understanding tasks, yet their practical deployment remains hindered by significant architectural heterogeneity across model families. This paper introduces UVLM (Universal Vision-Language Model Loader), a Google Colab-based framework that provides a unified interface for loading, configuring, and benchmarking multiple VLM architectures on custom image analysis tasks. UVLM currently supports two major model families -- LLaVA-NeXT and Qwen2.5-VL -- which differ fundamentally in their vision encoding, tokenization, and decoding strategies. The framework abstracts these differences behind a single inference function, enabling researchers to compare models using identical prompts and evaluation protocols. Key features include a multi-task prompt builder with support for four response types (numeric, category, boolean, text), a consensus validation mechanism based on majority voting across repeated inferences, a flexible token budget (up to 1,500 tokens) enabling users to design custom reasoning strategies through prompt engineering, and a built-in chain-of-thought reference mode for benchmarking. UVLM is designed for reproducibility, accessibility, and extensibility and as such is freely deployable on Google Colab using consumer-grade GPU resources. The paper also presents the first benchmarking of different VLMs on tasks of increasing reasoning complexity using a corpus of 120 street-view images.