1 paper · 1 filter
Yaman Umuroglu, Christoph Berganski, Felix Jentzsch +8
While neural network quantization effectively reduces the cost of matrix multiplications, aggressive quantization can expose non-matrix-multiply operations as significant performan…