🤖 AI Summary
To address the low fine-tuning efficiency and high GPU memory overhead of LLaMA-3.2-3B on medical chain-of-thought reasoning tasks under GPU and memory resource constraints, this paper proposes a two-stage parameter-efficient fine-tuning method integrating LoRA and QLoRA. By synergistically combining low-rank adaptation with 4-bit quantization, the approach enables lightweight adaptation of the full-parameter model on a single consumer-grade GPU (e.g., RTX 4090). Evaluated on standard medical reasoning benchmarks—MedQA-USMLE and PubMedQA—the method reduces GPU memory consumption by 60% and accelerates training by 2.3× compared to full-parameter fine-tuning, while preserving chain-of-thought coherence and medical factual accuracy (accuracy degradation <1.2%). This significantly enhances the practical deployability of large language models in resource-constrained clinical and medical AI settings.
📝 Abstract
Large Language Models (LLMs) such as GPT-4 and LLaMA have demonstrated remarkable reasoning abilities but require significant computational resources for fine-tuning. This paper presents a resource-efficient fine-tuning approach for LLaMA-3.2-3B to enhance medical chain-of-thought reasoning while operating under constrained GPU and memory settings. Using parameter-efficient tuning techniques such as LoRA and QLoRA, we adapt the base model on publicly available medical reasoning datasets. The model achieves improved reasoning coherence and factual accuracy while reducing memory usage by up to 60% compared to standard full fine-tuning. Experimental evaluation demonstrates that lightweight adaptations can retain strong reasoning capability in medical question-answering tasks. This work highlights practical strategies for deploying LLMs in low-resource research environments and provides insights into balancing efficiency and domain specialization for medical AI systems.