Cost-Effective Hyperparameter Optimization for Large Language Model Generation Inference

Chi Wang, Xueqing Liu, Ahmed Hassan Awadallah
Proceedings of the Second International Conference on Automated Machine Learning, PMLR 224:21/1-17, 2023.

Abstract

Large Language Models (LLMs) have sparked significant interest in their generative capabilities, leading to the development of various commercial applications. The high cost of using the models drives application builders to maximize the value of generation under a limited inference budget. This paper presents a study of optimizing inference hyperparameters such as the number of responses, temperature and max tokens, which significantly affects the utility/cost of text generation. We design a framework named EcoOptiGen which leverages economical hyperparameter optimization and cost-based pruning. Experiments with the GPT-3.5/GPT-4 models on a variety of tasks verify its effectiveness. EcoOptiGen is implemented in the “autogen” package of the FLAML library: \url{https://aka.ms/autogen}.

Cite this Paper

BibTeX
@InProceedings{pmlr-v224-wang23b, title = {Cost-Effective Hyperparameter Optimization for Large Language Model Generation Inference}, author = {Wang, Chi and Liu, Xueqing and Awadallah, Ahmed Hassan}, booktitle = {Proceedings of the Second International Conference on Automated Machine Learning}, pages = {21/1--17}, year = {2023}, editor = {Faust, Aleksandra and Garnett, Roman and White, Colin and Hutter, Frank and Gardner, Jacob R.}, volume = {224}, series = {Proceedings of Machine Learning Research}, month = {12--15 Nov}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v224/wang23b/wang23b.pdf}, url = {https://proceedings.mlr.press/v224/wang23b.html}, abstract = {Large Language Models (LLMs) have sparked significant interest in their generative capabilities, leading to the development of various commercial applications. The high cost of using the models drives application builders to maximize the value of generation under a limited inference budget. This paper presents a study of optimizing inference hyperparameters such as the number of responses, temperature and max tokens, which significantly affects the utility/cost of text generation. We design a framework named EcoOptiGen which leverages economical hyperparameter optimization and cost-based pruning. Experiments with the GPT-3.5/GPT-4 models on a variety of tasks verify its effectiveness. EcoOptiGen is implemented in the “autogen” package of the FLAML library: \url{https://aka.ms/autogen}.} }
Endnote
%0 Conference Paper %T Cost-Effective Hyperparameter Optimization for Large Language Model Generation Inference %A Chi Wang %A Xueqing Liu %A Ahmed Hassan Awadallah %B Proceedings of the Second International Conference on Automated Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Aleksandra Faust %E Roman Garnett %E Colin White %E Frank Hutter %E Jacob R. Gardner %F pmlr-v224-wang23b %I PMLR %P 21/1--17 %U https://proceedings.mlr.press/v224/wang23b.html %V 224 %X Large Language Models (LLMs) have sparked significant interest in their generative capabilities, leading to the development of various commercial applications. The high cost of using the models drives application builders to maximize the value of generation under a limited inference budget. This paper presents a study of optimizing inference hyperparameters such as the number of responses, temperature and max tokens, which significantly affects the utility/cost of text generation. We design a framework named EcoOptiGen which leverages economical hyperparameter optimization and cost-based pruning. Experiments with the GPT-3.5/GPT-4 models on a variety of tasks verify its effectiveness. EcoOptiGen is implemented in the “autogen” package of the FLAML library: \url{https://aka.ms/autogen}.
APA
Wang, C., Liu, X. & Awadallah, A.H.. (2023). Cost-Effective Hyperparameter Optimization for Large Language Model Generation Inference. Proceedings of the Second International Conference on Automated Machine Learning, in Proceedings of Machine Learning Research 224:21/1-17 Available from https://proceedings.mlr.press/v224/wang23b.html.

Related Material