Adapting organizational IT-solutions for local large language model support
Mäkelä, Tuukka (2025-06-04)
Adapting organizational IT-solutions for local large language model support
(04.06.2025)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
avoin
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe2025063075758
https://urn.fi/URN:NBN:fi-fe2025063075758
Tiivistelmä
This research aims to broaden understanding on the topic of localized large language models (LLMs) in organizational contexts. As a technology, these can offer a lot of advantages, that are suitable for the corporate world. Most notably, these offer significant benefits for controlling security related aspects – as such a system can be ran completely in-house, without any organizational and potentially sensitive data leaving the full control of a given organization.
Due to the demands of especially higher-end LLMs, most organizations are not prepared and ready to start supporting them on their pre-existing hardware structures. Whilst it’s normal for most organizations to have deployed computing solutions of some degree within their whole organization – due to the unique and ML-heavy focus of LLM-based workloads, these may not be compatible, forcing organizations to re-think their IT-solutions in order to begin utilizing localized LLMs.
This research has been completed on the basis of a few different methodologies and stages. Firstly, the driving factors and key trends impacting these developments were investigated with the aid of a systematic gray-literature review. After understanding these factors, the research aimed to understand the functionalities and requirements of a few key tiers of LLMs, through the completion of practical benchmark runs, as well as third-party benchmark data.
On the basis of these findings, the research is completed with a list of suggestions and factors of different tiers of LLM-systems. These include factors such as requirements and benefits an organization may expect to achieve from such systems. These solutions have been made on the principles of a design science-based artifact – giving anyone seeking to utilize these findings a solid foundation for understanding and completing such a project in a sensible and cost-effective manner. The artifact built within this research includes methods for planning such projects, and details about phases and tasks which should be completed to ensure a successful deployment takes place.
Due to the demands of especially higher-end LLMs, most organizations are not prepared and ready to start supporting them on their pre-existing hardware structures. Whilst it’s normal for most organizations to have deployed computing solutions of some degree within their whole organization – due to the unique and ML-heavy focus of LLM-based workloads, these may not be compatible, forcing organizations to re-think their IT-solutions in order to begin utilizing localized LLMs.
This research has been completed on the basis of a few different methodologies and stages. Firstly, the driving factors and key trends impacting these developments were investigated with the aid of a systematic gray-literature review. After understanding these factors, the research aimed to understand the functionalities and requirements of a few key tiers of LLMs, through the completion of practical benchmark runs, as well as third-party benchmark data.
On the basis of these findings, the research is completed with a list of suggestions and factors of different tiers of LLM-systems. These include factors such as requirements and benefits an organization may expect to achieve from such systems. These solutions have been made on the principles of a design science-based artifact – giving anyone seeking to utilize these findings a solid foundation for understanding and completing such a project in a sensible and cost-effective manner. The artifact built within this research includes methods for planning such projects, and details about phases and tasks which should be completed to ensure a successful deployment takes place.