{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:XFW6KW73WYUWT3RR3QVQ4TNZIQ","short_pith_number":"pith:XFW6KW73","schema_version":"1.0","canonical_sha256":"b96de55bfbb62969ee31dc2b0e4db944288bba7fe7ea519fb03e009b400641ae","source":{"kind":"arxiv","id":"2606.12854","version":1},"attestation_state":"computed","paper":{"title":"Small LLMs for Biomedical Claim Verification: Cost-Effective Fine-Tuning, Structural Dataset Shortcuts, and Cross-Domain Generalization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["q-bio.QM"],"primary_cat":"cs.CL","authors_text":"Gaurav Kumar","submitted_at":"2026-06-11T03:38:46Z","abstract_excerpt":"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 a"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2606.12854","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-06-11T03:38:46Z","cross_cats_sorted":["q-bio.QM"],"title_canon_sha256":"279845b34dd119ea56fe9e5cbe56b55e3bd2eeeddd969900539201c03d63fb45","abstract_canon_sha256":"3b6fbe7546b1e4afc27133f9a03305946b9ca8ad9cbe3623a6815e1801074517"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-12T01:08:54.556480Z","signature_b64":"Jf+o5Kxbnjl3ZnV8zaNJbWj1KRQHiiUTKlnW2yAMrsPiW72Qk8rzI/RaoDq+8yys/qlKfiFRI7EebmQ7BjyMBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b96de55bfbb62969ee31dc2b0e4db944288bba7fe7ea519fb03e009b400641ae","last_reissued_at":"2026-06-12T01:08:54.555594Z","signature_status":"signed_v1","first_computed_at":"2026-06-12T01:08:54.555594Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Small LLMs for Biomedical Claim Verification: Cost-Effective Fine-Tuning, Structural Dataset Shortcuts, and Cross-Domain Generalization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["q-bio.QM"],"primary_cat":"cs.CL","authors_text":"Gaurav Kumar","submitted_at":"2026-06-11T03:38:46Z","abstract_excerpt":"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 a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.12854","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2606.12854/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2606.12854","created_at":"2026-06-12T01:08:54.555734+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.12854v1","created_at":"2026-06-12T01:08:54.555734+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.12854","created_at":"2026-06-12T01:08:54.555734+00:00"},{"alias_kind":"pith_short_12","alias_value":"XFW6KW73WYUW","created_at":"2026-06-12T01:08:54.555734+00:00"},{"alias_kind":"pith_short_16","alias_value":"XFW6KW73WYUWT3RR","created_at":"2026-06-12T01:08:54.555734+00:00"},{"alias_kind":"pith_short_8","alias_value":"XFW6KW73","created_at":"2026-06-12T01:08:54.555734+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XFW6KW73WYUWT3RR3QVQ4TNZIQ","json":"https://pith.science/pith/XFW6KW73WYUWT3RR3QVQ4TNZIQ.json","graph_json":"https://pith.science/api/pith-number/XFW6KW73WYUWT3RR3QVQ4TNZIQ/graph.json","events_json":"https://pith.science/api/pith-number/XFW6KW73WYUWT3RR3QVQ4TNZIQ/events.json","paper":"https://pith.science/paper/XFW6KW73"},"agent_actions":{"view_html":"https://pith.science/pith/XFW6KW73WYUWT3RR3QVQ4TNZIQ","download_json":"https://pith.science/pith/XFW6KW73WYUWT3RR3QVQ4TNZIQ.json","view_paper":"https://pith.science/paper/XFW6KW73","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.12854&json=true","fetch_graph":"https://pith.science/api/pith-number/XFW6KW73WYUWT3RR3QVQ4TNZIQ/graph.json","fetch_events":"https://pith.science/api/pith-number/XFW6KW73WYUWT3RR3QVQ4TNZIQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XFW6KW73WYUWT3RR3QVQ4TNZIQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XFW6KW73WYUWT3RR3QVQ4TNZIQ/action/storage_attestation","attest_author":"https://pith.science/pith/XFW6KW73WYUWT3RR3QVQ4TNZIQ/action/author_attestation","sign_citation":"https://pith.science/pith/XFW6KW73WYUWT3RR3QVQ4TNZIQ/action/citation_signature","submit_replication":"https://pith.science/pith/XFW6KW73WYUWT3RR3QVQ4TNZIQ/action/replication_record"}},"created_at":"2026-06-12T01:08:54.555734+00:00","updated_at":"2026-06-12T01:08:54.555734+00:00"}