Abstract
The pervasive deployment of Large Language Models-LLMs in various sectorsoften neglects the nuanced requirements of individuals and small organizations,who benefit more from models precisely tailored to their specific businesscontexts rather than those with broadly superior general capabilities. Thiswork introduces \textbf{AnyTaskTune}, a novel fine-tuning methodology coined as\textbf{Task-Fine-Tune}, specifically developed to elevate model performance ona diverse array of domain-specific tasks. This method involves a meticulousprocess to identify and define targeted sub-tasks within a domain, followed bythe creation of specialized enhancement datasets for fine-tuning, therebyoptimizing task-specific model performance. We conducted comprehensivefine-tuning experiments not only in the legal domain for tasks such as keywordextraction and sentence prediction but across over twenty different sub-tasksderived from the domains of finance, healthcare, law, psychology, consumerservices, and human resources. To substantiate our approach and facilitatecommunity engagement, we will open-source these bilingual task datasets. Ourfindings demonstrate that models fine-tuned using the \textbf{Task-Fine-Tune}methodology not only achieve superior performance on these specific tasks butalso significantly outperform models with higher general capabilities in theirrespective domains. Our work is publicly available at\url{https://github.com/PandaVT/DataTager}.