Found 1 document(s)
Page 1 of 1
-
NVIDIA tensor cores have supported 2:4 semi-structured sparsity since Ampere, and the pattern composes with quantization. When the two are combined, the practical question is which step to take next: from a given precisi...