This repository is built to test LLMs for the MIDAS project. It focuses on running various open source LLMs that work best for epideimology data.
The epi_html and epi_pdf directories contain the code and results for evaluating the Epi dataset across various LLMs. These experiments were conducted with three different quantization levels and three different prompting strategies, across four LLMs.
Additionally, the epi_html_answerable_only and epi_pdf_answerable_only folders contain LLaMA-specific code that filters out responses such as “Unanswerable” and “System” from the generated answers.
