Is it necessary for commenting out the self.padding_mode != 'zeros'?

[youngstu]

Is it necessary for commenting out the self.padding_mode != 'zeros'? It won't report an error without making any modifications. Will it affect model accuracy. Thanks.

IMPORTANT: Please make sure to modify the site-packages/torch/nn/modules/conv.py file by commenting out the self.padding_mode != 'zeros' line to allow for replicated padding for ConvTranspose1d as shown https://github.com/NVIDIA/tacotron2/issues/182.

[youngstu]

It looks that ConvTranspose1d not used and ConvTranspose1d padding mode wouldn't affect the model accuracy.

https://github.com/Doubiiu/CodeTalker/blob/e687bbe64bb3553d5653041a9d80a62eae593ebf/config/BIWI/stage2.yaml#L25

    if args.quant_factor == 0:
        self.expander.append(nn.Sequential(
                    nn.Conv1d(size,dim,5,stride=1,padding=2,
                                padding_mode='replicate'),
                    nn.LeakyReLU(self.args.neg, True),
                    nn.InstanceNorm1d(dim, affine=args.INaffine)
                            ))
    else:
        self.expander.append(nn.Sequential(
                    nn.ConvTranspose1d(size,dim,5,stride=2,padding=2,
                                        output_padding=1,
                                        padding_mode='replicate'),
                    nn.LeakyReLU(self.args.neg, True),
                    nn.InstanceNorm1d(dim, affine=args.INaffine)
                            ))  

[aurelianocyp]

have you solved the problem? I also have this question because there are many self.padding_mode != 'zeros' , I don't konw comment out which one.

[Doubiiu]

Yes. You do not need to comment it out if you just use the proposed model without modifying the architecture (It is required to comment it out when modifying P in the main paper, indicating compression ratio of features in the temporal axis).