Currently, the ConvTranspose is only supported using the QNNPACK. The FBGEMM version is planned, but there is no specific date specified for it. Meanwhile, you have two options for the eager mode: replace the ConvTranspose: 1) Replace the instances of the ConvTranspose with dequant->ConvTranspose->quant construct 2) Set the torch.backends.quantized.engine = 'qnnpack' before running your model. You also might need to set the qconfig = torch.quantization.get_default_qconfig('qnnpack') or qconfig = torch.quantization.get_default_qat_qconfig('qnnpack')