POINTS: Improving Your Vision-language Model with Affordable Strategies

TL;DR

This paper introduces affordable, effective strategies for improving vision-language models, including data filtering, model ablations, and model soup techniques, resulting in a competitive 9B parameter model.

Contribution

The paper presents a robust baseline, data filtering with perplexity, and model soup methods to enhance vision-language models efficiently.

Findings

Achieved competitive performance with a 9B parameter model.

Curated a 1M dataset using perplexity filtering for pre-training.

Used model soup to improve fine-tuning results.

Abstract

In recent years, vision-language models have made significant strides, excelling in tasks like optical character recognition and geometric problem-solving. However, several critical issues remain: 1) Proprietary models often lack transparency about their architectures, while open-source models need more detailed ablations of their training strategies. 2) Pre-training data in open-source works is under-explored, with datasets added empirically, making the process cumbersome. 3) Fine-tuning often focuses on adding datasets, leading to diminishing returns. To address these issues, we propose the following contributions: 1) We trained a robust baseline model using the latest advancements in vision-language models, introducing effective improvements and conducting comprehensive ablation and validation for each technique. 2) Inspired by recent work on large language models, we filtered pre-training data using perplexity, selecting the lowest perplexity data for training. This approach allowed us to train on a curated 1M dataset, achieving competitive performance. 3) During visual instruction tuning, we used model soup on different datasets when adding more datasets yielded marginal improvements. These innovations resulted in a 9B parameter model that performs competitively with state-of-the-art models. Our strategies are efficient and lightweight, making them easily adoptable by the community.