BitNet b1.58 2B4T Technical Report

Benchmark Model Rank Results
instruction-following-on-mtbenchBitNet b1.58 2B4T#16Score (1-10): 5.85
question-answering-on-triviaqaBitNet b1.58 2B4T#45EM: 33.57
multi-task-language-understanding-on-mmluBitNet b1.58 2B4T#61Average (%): 53.17
common-sense-reasoning-on-winograndeBitNet b1.58 2B4T#70Accuracy: 71.90
sentence-completion-on-hellaswagBitNet b1.58 2B4T#115Accuracy: 68.44
mathematical-reasoning-on-math-500BitNet b1.58 2B4T#126Accuracy: 43.40
arithmetic-reasoning-on-gsm8kBitNet b1.58 2B4T#208Accuracy: 58.38