I’ve followed the tutorial on quantizing using my own mxnet model and load the compiled module in my c++ app via the GraphRuntime (no code changes from non-quantized).
I’ve Inspected the outputs of the net and they are surely wrong.
I’ve tried ignorantly fiddling with the qconfig parameters, but no luck. Maybe I’m hitting an unsupported scenario?
What are the suggestions on troubleshooting this?
Thanks for any help.