QLoRA fine-tuning and evaluation framework for medical reasoning LLMs using Qwen2.5-3B with CoT vs No-CoT comparison.

A framework for fine-tuning and evaluating medical reasoning LLMs using QLoRA on Qwen2.5-3B, comparing chain-of-thought prompting versus no chain-of-thought.

What happened

The repository provides tools for QLoRA fine-tuning of the Qwen2.5-3B model specifically targeting medical reasoning tasks, including a systematic evaluation of chain-of-thought (CoT) versus no-CoT prompting methods.

Why it matters

This work advances the adaptation of efficient fine-tuning methods for specialized medical LLMs, enabling exploration of reasoning techniques critical for complex domain-specific question answering.