HAZMAT Large Language Model Fine-Tuning and Evaluation Datasets

This repository contains the scientific data generated during the evaluation and fine-tuning of offline, edge-deployed Large Language Models (LLMs) for Hazardous Materials (HAZMAT) emergency response. The goal of this research was to determine if sub-4GB open-weights models could be fine-tuned to act as reliable, offline safety advisors for field workers using local inference, benchmarking their performance against commercial off-the-shelf (COTS) cloud models.