arXiv cs.CL· Rezwan Islam Salvi, Saiyma Sittul Muna, Mushfiqur Rahman Mushfique·· 10 小时前AI 评分31
HandsOnAid：面向应急救援场景的离线急救助手构建与评测
HandsOnAid: Building and Evaluating an Offline First-Aid Assistant for Emergency Response Scenarios
AI 导读
研究者构建了 HandsOnAid，一个基于 4-bit Gemma 2B、用 QLoRA 在 5,550 条急救问答对 FirstAidQA 上微调的离线急救助手。微调使平均分较基座模型提升 +1.08 至 2.34，最佳配置为带检索的 4-bit 微调模型，得分 2.51/5。在无网络的 Pixel 7 Pro 上，该模型平均 6.86 秒输出首段可显示文本、14.88 秒生成完整回答。
正文 · 原文
Abstract:Emergencies can occur where there is no phone signal and no ambulance nearby. A small language model on a phone can be helpful in such situations, but its advice needs to be safe and accurate. We built HandsOnAid, a 4-bit Gemma 2B model fine-tuned with QLoRA on FirstAidQA dataset, a corpus of 5,550 first-aid question-answer pairs. We compared seven configurations, including the base, 4-bit and 8-bit fine-tuned models, and versions with a length floor, safety gate, or retrieval. We used a 41-pair question-answer bank with medically reviewed references that followed ANZCOR guidelines, and a strict 0-5 score rubric, with twelve safety overrides. Four frontier LLM judges (GPT-5.6 Sol, Claude Opus 5, DeepSeek V4 Pro, GLM-5.3) blindly scored every answer, with 45 hidden controls used to check judge reliability. Fine-tuning improved the mean score by +1.08 over the base model to 2.34. The best configuration scored 2.51 out of 5 which was the fine-tuned 4-bit model with retrieval. A medical professional also conducted a blind evaluation of 40 answers across the base, the 4-bit fine-tuned, and the retrieval configurations, and the scores were similar to the LLM judges' average scores. A keyword audit found limited coverage of several details of ANZCOR guidelines in the training corpus, that are required by the rubric, which suggests that missing information in the training data could be a cause of the low scores. On a smartphone device (Pixel 7 Pro) with six CPU threads and no network connection, the 4-bit fine-tuned model produced its first displayable text in 6.86 seconds and a complete answer in 14.88 seconds on average over ten questions.
| Comments: | 12 pages, 2 figures, 10 tables |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2611.14059 [cs.CL] |
| (or arXiv:2611.14059v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.14059 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Rezwan Islam Salvi [view email]
[v1]
Tue, 10 Nov 2026 13:25:28 UTC (475 KB)
来源：arXiv cs.CL · arxiv.org