What is Structured sparsity?
Sparsity in which zero weights follow a fixed pattern; 2:4 sparsity, with two zeros in every four weights, lets matrix units compute about twice as fast.
Field: AI Hardware & Software Stack
Sparsity in which zero weights follow a fixed pattern; 2:4 sparsity, with two zeros in every four weights, lets matrix units compute about twice as fast.
Field: AI Hardware & Software Stack
GPU AI compiler AI inference All-reduce Arithmetic intensity Attention mechanism Backpropagation Batch size Computational graph Data parallelism