Quantization doesn't set the right backend

After quantization (with USE_CUDA = True) and trying to run an inference. I get an error CUDNN CUDNN_STATUS_BAD_PARAM (2000).

This error seems to occure when trying to get the weight tensor of a Convolution.