Dynamic Linear Attention

Benchmark Model Rank Results
common-sense-reasoning-on-arc-easyMamba-2-780M w/ DLA–Accuracy: 48.1
common-sense-reasoning-on-winograndeGated DeltaNet-1.3B w/ DLA–Accuracy: 52.4
question-answering-on-natural-questionsGated DeltaNet-1.3B w/ DLA–EM: 12.1
question-answering-on-piqaMamba-2-780M w/ DLA–Accuracy: 63.7
sentence-completion-on-hellaswagMamba-2-780M w/ DLA–Accuracy: 30.8