Software and Dataset For Fine-Tuning a local LLM for Thermo-Electric Generators with QLoRA: from generalist to specialist

This document provides a comprehensive guide to the software and datasets contained inthis repository. These resources were developed for the research presented in the article: "[Ar-ticle Title Fine-Tuning a local LLM for Thermo-Electric Generators with QLoRA: from gener-alist to specialist:;Preprinthttps://www.preprints.org/manuscript/202511.1348 ]". This work presents a complete pipeline for transforming general-purposeLarge Language Models (LLMs) into specialized technical assistants for thermoelectric genera-tor (TEG) applications. Using QLoRA (Quantized Low-Rank Adaptation), we eciently ne-tuned two open-source modelsJanV1-expert-TEG and Qwen3-4B-thinking-2507-TEG on acurated dataset of 202 question-answer pairs covering thermoelectric materials, device physics,performance optimization, and engineering applications. The methodology enables special-ization without full model retraining, signicantly reducing computational requirements whilemaintaining technical accuracy. We introduce a novel evaluation framework combining humanexpert review with LLM-as-a-Judge scoring using state-of-the-art models (GPT-4, Gemini 1.5Pro). The resulting specialist models demonstrate enhanced performance on technical queries,achieving higher relevance and accuracy scores compared to their base counterparts. All re-sources, including trained adapters, merged models, training scripts, and evaluation datasets,are provided for reproducibility and further research in domain-specic AI applications forrenewable energy technologies.本文件为本仓库收录的软件与数据集提供全面使用指南。本项目配套资源均为支持下述研究论文而开发：《基于QLoRA将本地大语言模型微调为热电发电机专用助手：从通用到专精》[预印本，https://www.preprints.org/manuscript/202511.1348]。本研究提出一套完整流程，可将通用型大语言模型（Large Language Model，LLM）转化为适用于热电发电机（Thermoelectric Generator，TEG）场景的专用技术助手。本研究采用QLoRA（量化低秩适配，Quantized Low-Rank Adaptation）技术，针对202条经精选的问答对数据集，高效微调了两款开源模型：JanV1-expert-TEG与Qwen3-4B-thinking-2507-TEG。该数据集涵盖热电材料、器件物理、性能优化及工程应用等领域内容。该微调方法无需对模型进行全量重新训练即可实现领域专精化，在保障技术准确性的同时，大幅降低了计算资源需求。本研究提出一种全新的评估框架，结合人工专家评审与大语言模型作为评判者（LLM-as-a-Judge）的评分机制，采用当前前沿模型（GPT-4、Gemini 1.5 Pro）完成评分。经微调得到的专用模型在技术类查询任务中表现更优，相较于其基础模型版本，在相关性与准确性评分上均取得了更优结果。本项目提供全部配套资源，包括训练得到的适配器、融合模型、训练脚本及评估数据集，以支持可再生能源技术领域专用人工智能应用的可复现研究与后续拓展工作。