Automated Generation of Discharge Summaries: Leveraging Large Language Models With Clinical Data

This repository contains the codebase used for the automated generation of discharge summaries, leveraging LLaMA3 and German clinical data.

The full-text paper detailing the methodology and results can be found at [link to paper/tbd].

Code Components

To create the prompt we used create_prompt.py and to run the model run_model.py. To compute our quantitative metrics we used compute_bertscore.py and compute_rouge.py. To calculate BERTScore and ROUGE means and standard deviation as well as the overall generation time we used compute_stats.py. The central script to manage the entire workflow can be found in main.py.

Citation

tbd