Gated Recurrent Transformers: Expressive Depth through Recurrent Modulation

Benchmark Model Rank Results
zero-shot-question-answering-on-openbookqaGRT Large (isoParams, zero-shot)#22Accuracy: 30.2
multiple-choice-question-answering-mcqa-on-arc-eGRT Large (isoParams, zero-shot)#24Accuracy: 45.54
multiple-choice-question-answering-mcqa-on-arc-cGRT Large (isoFLOPs, zero-shot)#29Accuracy: 25.43
language-modelling-on-lambadaGRT Large (isoParams, zero-shot)#53Accuracy: 45.27
question-answering-on-boolqGRT Large (isoFLOPs, zero-shot)#102Accuracy: 60.18
common-sense-reasoning-on-winograndeGRT Large (isoParams, zero-shot)#155Accuracy: 53.28
question-answering-on-piqaGRT Large (isoParams, zero-shot)#159Accuracy: 66.05
sentence-completion-on-hellaswagGRT Large (isoParams, zero-shot)#198Accuracy: 41.52