6G Master Thesis: Resource Efficient Large Language Models at the Edge

Purpose and goal

The overall aim of the project is to develop computationally efficient large language models (LLMs) for use in resource-constrained 6G-edge environments. The goal of this thesis is therefore to explore energy-efficient transformer-based language models by utilizing advanced techniques such as knowledge distillation and model quantization.

Expected effects and result

Expected results from the project include a comparison based on numerical experiments between a new alternative architecture and conventional transformer models. In a larger perspective, the project has potential to contribute to improved computational efficiency and more sustainable AI services in 6G edge environments.

Planned approach and implementation

The project will be implemented in several steps. First, the student conducts a literature study to understand existing techniques and models. Then, the student conducts practical experiments to train and optimize new computationally efficient language models. These are compared with conventional transformer models through numerical experiments. Finally, the results are analyzed and documented, and the methods and results are made available as open source code.