Large language models (LLMs) excel in general tasks but often struggle with domain-specific applications due to limited specialized knowledge. Moreover, in practical applications, it is often necessary to handle multiple tasks within a specific domain. However, existing methods for applying large language models are predominantly focused on single-task optimization. To address these problems, we propose a domain-adaptive multi-task fine-tuning method that enables LLMs to handle diverse tasks within a domain. Our approach introduces two lightweight modules: domain adapters for capturing shared domain knowledge and a task embedding layer for distinguishing task-specific differences, both with low computational overhead. Additionally, we design a unified task conversion template to standardize diverse tasks, facilitating efficient multi-task learning. Experiments in the medical domain, including ChemProt and Symptom-to-diagnosis tasks, demonstrate that our method outperforms existing fine-tuning techniques, achieving superior performance and remaining efficient.