CiteCheck verifier (DeBERTa-v3-base fine-tuned on SciFact)

Judges whether evidence sentences from a paper's abstract SUPPORT, CONTRADICT or are NEUTRAL to a scientific claim. Fine-tuned on SciFact train (CC BY 4.0) minus 200 claims held out for threshold tuning; see train_log.json. SciFact dev, end to end in CiteCheck: macro-F1 0.693 (95% CI 0.636-0.749), false-SUPPORT rate 0.068. Biomedical abstracts only; not validated on other fields. Course project (UMD DATA/MSML 641).

Downloads last month
14
Safetensors
Model size
0.2B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for hassanedwan/citecheck-nli-ft-base

Finetuned
(9)
this model

Dataset used to train hassanedwan/citecheck-nli-ft-base