GLEAN: Grounded Lightweight Evaluation Anchors for Contamination-Aware Tabular Reasoning

Benchmark Model Rank Results
table-based-fact-verification-on-tabfactDeBERTa-v3 (GLEAN protocol)–Test: 57.4
table-claim-verification-on-scitabDeBERTa-v3 (GLEAN protocol)–Macro-F1 (3-class): 0.1970
table-question-answering-on-tablebenchTAPAS-base–EM: 0.093F1: 0.119
table-question-answering-on-tablebenchTAPAS-large–EM: 0.090F1: 0.121
table-question-answering-on-tablebenchTAPEX–EM: 0.105F1: 0.122