An Empirical Study on Hyperparameter Optimization for Fine-Tuning Pre-trained Language Models

TL;DR

This study evaluates hyperparameter optimization methods for fine-tuning pre-trained language models, revealing their limitations and proposing strategies to improve their effectiveness, with insights into overfitting and search space configurations.

Contribution

It systematically analyzes HPO performance on language models, identifies failure causes, and offers troubleshooting strategies to enhance HPO success in NLP tasks.

Findings

HPO often fails to outperform grid search within the same time budget

Insufficient time budget and overfitting are key failure reasons

Proper search space and time budget adjustments can improve HPO outcomes

Abstract

The performance of fine-tuning pre-trained language models largely depends on the hyperparameter configuration. In this paper, we investigate the performance of modern hyperparameter optimization methods (HPO) on fine-tuning pre-trained language models. First, we study and report three HPO algorithms' performances on fine-tuning two state-of-the-art language models on the GLUE dataset. We find that using the same time budget, HPO often fails to outperform grid search due to two reasons: insufficient time budget and overfitting. We propose two general strategies and an experimental procedure to systematically troubleshoot HPO's failure cases. By applying the procedure, we observe that HPO can succeed with more appropriate settings in the search space and time budget; however, in certain cases overfitting remains. Finally, we make suggestions for future work. Our implementation can be found in https://github.com/microsoft/FLAML/tree/main/flaml/nlp/.