The rapid integration of Large Language Models (LLMs) into various industries presents both revolutionary opportunities and unique challenges. This research aims to establish a scalable and efficient framework for LLM customization, exploring how DevOps practices should be adapted to meet the specific demands of LLM customization. By integrating ontologies, knowledge maps, and prompt engineering into the DevOps pipeline, we propose a robust framework that enhances continuous learning, seamless deployment, and rigorous version control of LLMs. This methodology is demonstrated through the development of a domain-specific chatbot for the agricultural sector, utilizing heterogeneous data to deliver actionable insights. The proposed methodology, so called DOLLmC, not only addresses the immediate challenges of LLM customization but also promotes scalability and operational efficiency. However, the methodology's primary limitation lies in the need for extensive testing, validation, and broader adoption across different domains.
@article{arxiv.2405.11581,
title = {DOLLmC: DevOps for Large Language model Customization},
author = {Panos Fitsilis and Vyron Damasiotis and Vasileios Kyriatzis and Paraskevi Tsoutsa},
journal= {arXiv preprint arXiv:2405.11581},
year = {2024}
}