Sequential Large Language Model-Based Hyper-parameter Optimization

TL;DR

This paper presents SLLMBO, a novel framework using large language models for hyperparameter optimization, incorporating dynamic search space adaptation and a new LLM-TPE sampler, outperforming traditional methods on multiple tasks.

Contribution

Introduces SLLMBO, the first framework benchmarking diverse LLMs for HPO, with a novel LLM-TPE sampler balancing exploration and exploitation effectively.

Findings

LLM-TPE outperforms fully LLM-based methods on 9 of 14 tasks.

SLLMBO achieves more robust optimization than traditional Bayesian methods.

The framework reduces API costs and mitigates premature early stopping.

Abstract

This study introduces SLLMBO, an innovative framework leveraging large language models (LLMs) for hyperparameter optimization (HPO), incorporating dynamic search space adaptability, enhanced parameter space exploitation, and a novel LLM-tree-structured parzen estimator (LLM-TPE) sampler. By addressing limitations in recent fully LLM-based methods and traditional bayesian optimization (BO), SLLMBO achieves more robust optimization. This comprehensive benchmarking evaluates multiple LLMs, including GPT-3.5-Turbo, GPT-4o, Claude-Sonnet-3.5, and Gemini-1.5-Flash, extending prior work and establishing SLLMBO as the first framework to benchmark a diverse set of LLMs for HPO. By integrating LLMs' established strengths in parameter initialization with the exploitation abilities demonstrated in this study, alongside TPE's exploration capabilities, the LLM-TPE sampler achieves a balanced exploration-exploitation trade-off, reduces API costs, and mitigates premature early stoppings for more effective parameter searches. Across 14 tabular tasks in classification and regression, the LLM-TPE sampler outperformed fully LLM-based methods and achieved superior results over BO methods in 9 tasks. Testing early stopping in budget-constrained scenarios demonstrated competitive performance, indicating that LLM-based methods generally benefit from extended iterations for optimal results. This work lays the foundation for future research exploring open-source LLMs, reproducibility of LLM results in HPO, and benchmarking SLLMBO on complex datasets, such as image classification, segmentation, and machine translation.