Quick-Tune-Tool: A Practical Tool and its User Guide for Automatically Finetuning Pretrained Models

Ivo Rapant, Lennart Purucker, Fabio Ferreira, Sebastian Pineda Arango, Arlind Kadra, Josif Grabocka, Frank Hutter

Keywords: Pretrained Models, Finetuning, Automated Model Selection, Quick-Tune-Tool, Image Classification, Machine Learning, Hyperparameter Optimization, Quick-Tune, Vision, Neural Networks, Meta-Learning, Search Space, Objective Function, AutoML

TL;DR: Quick-Tune-Tool is an automated system designed to simplify the selection and finetuning of pretrained models.

Abstract: Pretrained models have become essential tools for machine learning practitioners across various domains including image classification, segmentation, and natural language processing. However, the complexity of selecting the appropriate pretrained model and finetuning strategy remains a significant challenge. In this paper, we present Quick-Tune-Tool, an automated solution to guide practitioners in selecting and finetuning pretrained models. Leveraging the Quick-Tune algorithm, Quick-Tune-Tool abstracts intricate research-level code into a user-friendly tool. Our contributions include the release of Quick-Tune-Tool, a detailed architectural overview, a user guide for image classification, and empirical evaluations. In experiments on four vision dataset, our results underscore the effectiveness and practicality of Quick-Tune-Tool for automating model selection and finetuning.

Submission Checklist: Yes

Broader Impact Statement: Yes

Paper Availability And License: Yes

Code Of Conduct: Yes

Optional Meta-Data For Green-AutoML: All questions below on environmental impact are optional.

Submission Number: 12