Title:	DL4NLP: Deep Learning for Natural Language Processing
DNr:	Berzelius-2025-330
Project Type:	LiU Berzelius
Principal Investigator:	Marco Kuhlmann <marco.kuhlmann@liu.se>
Affiliation:	Linköpings universitet
Duration:	2025-10-01 – 2026-04-01
Classification:	10208
Homepage:	https://www.ida.liu.se/divisions/aiics/nlp/
Keywords:

Abstract

The Natural Language Processing (NLP) group at Linköping University is working on developing techniques for fine-tuning and analyzing large language models, a detailed introduction of which could be found in the next section. Given our recent utilization patterns and the influx of new members into our group, it's important to address our resource needs for the foreseeable future. We've nearly exhausted our current quota outside of summer months, and with numerous new members joining our group and getting started with their research activities, our demand for resources is only going to increase. While it's difficult to predict the exact amount needed, we require an increased quota to 17950 hours per month. Although GPU usage dipped during the summer months, the start of new projects this semester, coupled with the expanding resource requirements within the NLP research community, underscores the necessity of increasing our current allocation. A growing part of our research agenda focuses on modularization strategies and cross-lingual dynamics, where controlled experiments often require training models from scratch rather than relying on industry-provided checkpoints. Since these experiments typically involve training on multiple languages simultaneously, the computational demand scales steeply with each additional language. The trend of increasingly larger models and datasets in NLP research compounds this demand. Our experience with queuing issues, particularly during critical periods such as the upcoming December and January conference deadlines, highlights the importance of having access to the requested GPU hours. Ensuring continuity in our research momentum and productivity depends on having access to adequate resources to meet our evolving needs.