A new benchmark study evaluated eight open-source small language models (SLMs) for emergency department (ED) decision support, comparing them against commercial models like Claude Haiku 4.5 and Claude Sonnet 4.5. The research found that SLMs fine-tuned with Low-Rank Adaptation (LoRA) outperformed the commercial baselines in predicting triage levels and recommending specialist referrals. While diagnosis prediction remains a challenge for open-source SLMs, the fine-tuned models demonstrated an ability to identify high-severity patients missed by commercial alternatives, suggesting their potential for clinically competitive performance in local ED settings. AI
IMPACT Demonstrates potential for privacy-preserving, locally deployable LLMs in critical healthcare settings.
RANK_REASON Academic paper detailing a systematic benchmark of fine-tuning strategies for small language models. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- Claude Haiku 4.5
- Claude Sonnet 4.5
- DagsHub
- Gotit.pub
- Hugging Face
- LoRA
- MIMIC-IV-ED
- ScienceCast
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
