{"slug": "small-llms-for-biomedical-claim-verification-cost-effective-fine-tuning-dataset", "title": "Small LLMs for Biomedical Claim Verification: Cost-Effective Fine-Tuning, Structural Dataset Shortcuts, and Cross-Domain Generalization", "summary": "Researchers fine-tuned three small language models—Phi-3-mini, Qwen2.5-3B, and Mistral-7B—using QLoRA on biomedical claim verification datasets, finding that Mistral-7B outperformed GPT-4o and GPT-5 by up to 12% F1 at a fraction of the cost with only 1,008 training examples. The study identified a previously unreported structural artifact in the SciFact dataset that inflated in-domain scores, and demonstrated that training on structurally sound data enabled robust cross-domain generalization. The findings suggest that cost-effective fine-tuning of small models can match or exceed large proprietary systems for biomedical claim verification, with all code and adapter checkpoints to be released.", "body_md": "arXiv:2606.12854v1 Announce Type: new\nAbstract: Large Language Models such as GPT-4o and GPT-5 achieve strong zero-shot performance on biomedical claim verification, but cost and opacity limit scalable use. We fine-tune three small LLMs: Phi-3-mini (3.8B), Qwen2.5-3B, and Mistral-7B, via QLoRA on SciFact and HealthVer, providing the first study of QLoRA models against GPT-4o and fine-tuned BioLinkBERT encoders. Mistral-7B QLoRA surpasses both GPT-4o and GPT-5 (up to 12% F1 gain) at a fractional cost using just 1,008 training examples. We conduct extensive in-domain and cross-domain evaluation: models trained on SciFact tested on HealthVer and vice versa, at matched sizes to isolate dataset structure from data quantity. We identify a previously unreported structural artifact in SciFact that inflates in-domain scores, and show through bidirectional out-of-domain evaluation that training on structurally sound data enables robust cross-domain transfer. We plan to release all code and adapter checkpoints.", "url": "https://wpnews.pro/news/small-llms-for-biomedical-claim-verification-cost-effective-fine-tuning-dataset", "canonical_source": "https://arxiv.org/abs/2606.12854", "published_at": "2026-06-12 04:00:00+00:00", "updated_at": "2026-06-12 04:56:25.639635+00:00", "lang": "en", "topics": ["large-language-models", "natural-language-processing", "artificial-intelligence", "machine-learning", "ai-research"], "entities": ["GPT-4o", "GPT-5", "Phi-3-mini", "Qwen2.5-3B", "Mistral-7B", "SciFact", "HealthVer", "BioLinkBERT"], "alternates": {"html": "https://wpnews.pro/news/small-llms-for-biomedical-claim-verification-cost-effective-fine-tuning-dataset", "markdown": "https://wpnews.pro/news/small-llms-for-biomedical-claim-verification-cost-effective-fine-tuning-dataset.md", "text": "https://wpnews.pro/news/small-llms-for-biomedical-claim-verification-cost-effective-fine-tuning-dataset.txt", "jsonld": "https://wpnews.pro/news/small-llms-for-biomedical-claim-verification-cost-effective-fine-tuning-dataset.jsonld"}}