Error in trans_conv2d when model is reparametrized
rameshamurtaza opened this issue · comments
conv_transpose2d(): argument 'groups' (position 7) must be int, not tuple
I used this command to convert my model to bnn (const_bnn_prior_parameters) and now getting this error
I believe I found the error. In line 812 of bayesian_torch/layers/variational_layers/conv_variational.py
the F.conv_transpose2d
function is being called as:
out = F.conv_transpose2d(input, weight, bias, self.stride,
self.padding, self.output_padding,
self.dilation, self.groups)
i.e. the order is dilation then groups, whereas as seen in https://pytorch.org/docs/stable/generated/torch.nn.functional.conv_transpose2d.html , the order should be groups then dilation:
out = F.conv_transpose2d(input, weight, bias, self.stride,
self.padding, self.output_padding,
self.groups, self.dilation)
Fixed. commit 97ba16a