Knowledge-in-Context: Towards Knowledgeable Semi-Parametric Language Models

Benchmark Model Rank Results
common-sense-reasoning-on-winograndeKiC-770MAccuracy: 55.30
coreference-resolution-on-winograd-schemaKiC-770MAccuracy: 65.40
natural-language-inference-on-anli-testKiC-770MA1: 36.30A2: 35.00A3: 37.60
natural-language-inference-on-rteKiC-770MAccuracy: 74.00
question-answering-on-copaKiC-770MAccuracy: 85.30
question-answering-on-storyclozeKiC-770MAccuracy: 94.40
sentence-completion-on-hellaswagKiC-770MAccuracy: 29.6
word-sense-disambiguation-on-words-in-contextKiC-770MAccuracy: 52.40