Automatic Evaluation Code for Measuring Dialogue Generation Model Performance

Leveraging Large Language Models for text generation and dialogue systems, with advanced evaluation and retrieval techniques.

The project investigates follow-up question generation in medical dialogues by fine-tuning large language models on synthetic, ICF-grounded data. This repository contains the synthetic training data generation script and training data, evaluation data, evaluation scripts, and experiment outputs.

Fine-tuning and evaluating large language models (LLaMA-2 7B & Mistral 7B) using LoRA on the Empathetic Dialogues dataset — complete with quantization, auto-resume, generation, and metrics visualization.