Quantization error about _conv_transpose2d

quantization
1

Hey,Guys!I had tried to use dequant and quant to ignore the convtranspose2d like this:

class Conv2dTranspose(nn.Module):
    def __init__(self, cin, cout, kernel_size, stride, padding, output_padding=0, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.conv_block = nn.Sequential(
                            nn.ConvTranspose2d(cin, cout, kernel_size, stride, padding, output_padding),
                            nn.BatchNorm2d(cout),
                            nn.ReLU()
                            )

        #self.act = nn.ReLU()
        self.quant = torch.ao.quantization.QuantStub()
        self.dequant = torch.ao.quantization.DeQuantStub()

        def get_middle_todequant(module,args):
            output = self.dequant(args[0])
            module.qconfig = None
            return (output)

        def get_middle_toquant(module,args,output):
            output = self.quant(output)
            return output
        self.handle = self.conv_block[0].register_forward_pre_hook(get_middle_todequant)
        self.handle2 = self.conv_block[0].register_forward_hook(get_middle_toquant)
    def forward(self, x):
        out = self.conv_block(x)
        # out = self.dequant(out)
        return out

but still has an error like this:

NotImplementedError: Could not run 'aten::slow_conv_transpose2d.out' with arguments from the 'QuantizedCPU' backend. This could be because the operator doesn't exist for this backend, or was omitted during the selective/custom build process (if using custom build).

My torch version is 2.0.1+cu17,Is there any problems?

2

either your stubs aren’t getting converted to actual quant/dequant or your prehook isn’t actually updating the output (i think you might need to either modify args directly or not use hooks for this)

3

Thank you for providing the idea. In fact, the hook did not work properly. I have modified the code to ensure that the input and output of the ConvTranspose2d layer use quant and dequant