Training large language models (LLMs) at large scale is computationally nontrivial task. The choice of programming language introduces restrictions, even with using very well optimized libraries (e.g. Pytorch). In this thesis, the student will focus on using a pre-trained LLM over knowledge graph to predict missing links. To this end, the student will closely work on llm.c. Please refer to Learning continuous representations for Knowledge Graphs for details about knowledge graphs and predicting missing links.
In case you have further questions, feel free to contact Caglar Demir.