MPolaris

Quantification is usually divided into two modes: per-channel or per-tensor. For per-channel zero_point and scale are 1d array and it's length same as tensor channel. For per-tensor zero_point and scale...

In QNN-Frontent, the zero_point parameter reading method seems to be incorrect. In the '_qnn_conv2d_legalize_cuda' function, data of type uint8 will be shift, but only when zero_point is a scalar, i.e....