Rename attribute in SamePadConvTranspose to better match checkpoint (#207)

Summary:
Simple change in `SamePadconvTranspose3d` from `self.conv` to `self.convt` for an easier mapping to MUGEN's checkpoint for VQVAE.

Pull Request resolved: #207

Test Plan: `pytest test -vv`

Reviewed By: langong347

Differential Revision: D38178382

Pulled By: RdoubleA

fbshipit-source-id: 928ac24b946c42ce8404b9abbc985c10b9ceb65f

Author: RdoubleA

torchmultimodal/modules/layers/conv.py

@@ -108,7 +108,7 @@ def __init__(
                 use ConvTranspose3d directly for custom padding"
             )
 
-        self.conv = nn.ConvTranspose3d(
+        self.convt = nn.ConvTranspose3d(
             in_channels, out_channels, kernel_size, stride=stride, bias=bias, **kwargs
         )
 
@@ -118,10 +118,10 @@ def forward(self, x: Tensor) -> Tensor:
             self.pad_input = calculate_same_padding(
                 self.kernel_size, self.stride, x.shape[2:]
             )
-            self.conv.padding, self.conv.output_padding = calculate_transpose_padding(
+            self.convt.padding, self.convt.output_padding = calculate_transpose_padding(
                 self.kernel_size, self.stride, x.shape[2:], self.pad_input
             )
-        return self.conv(F.pad(x, self.pad_input))
+        return self.convt(F.pad(x, self.pad_input))
 
 
 def calculate_same_padding(