Custom static PTQ patterns


I am new to PyTorch. How do I bring in custom quantization pattern in pytorch?
For example: all the tensors going into concat needs to have the same scale and zp and the concat op itself. How do I bring this about with qnnpack backend?

Hi hh1234,

Have you already taken a look at our tutorials for quantization in eager mode and fx? Also FYI I think that quantized concat is already supported.