Cost-Effective Hyperparameter Optimization for Large Language Model Generation Inference
This work addresses cost reduction for application builders using LLMs, but it is incremental as it applies existing hyperparameter optimization techniques to a new context.
The paper tackles the problem of high inference costs for large language models by optimizing hyperparameters like number of responses and temperature to maximize utility under a budget, resulting in a framework called EcoOptiGen that shows effectiveness in experiments with GPT-3.5/GPT-4 models.
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}.