Large Language Model Operations (LLMOps): Definition, Challenges, and Lifecycle Management
Josu Díaz-de-Arcaya, Juan López-de-Armentia, Raúl Miñón, Iker Lasa Ojanguren, Ana I. Torre-Bastida
Abstract
Numerous studies explore the prospects presented by the recent upsurge of large language models. The usage of LLMs in production environments poses challenges that highlight the limitations of methodologies such as MLOps, and further investigation in this field is required. To this end, a new methodology, coined large language model operations (LLMOps), has arisen to address the particularities of LLMs. This term is so recent that the scientific literature has not yet agreed on a common definition for it, and the use of non-peer reviewed studies becomes a must. In this research, we review the current literature in the field to shed light on the adoption of LLMOps to drive innovation and efficiency in deploying large language models in real-world applications. To this end, three research questions are used to guide the contribution to the scientific literature with a unified definition of LLMOps, the challenges posed by LLMs that require the need for this new methodology, and to outline the key stages of LLMOps and their particularities that must be considered.