torch.nn.ConvTranspose2d fails with constant output_size

[elad-c]

Issue Type

Bug

Source

source

MCT Version

nightly

OS Platform and Distribution

No response

Python version

No response

Describe the issue

MCT builds the quantized model incorrectly. The quantized model output shape is different between original and quantized models.

Expected behaviour

No response

Code to reproduce the issue

class Model(nn.Module):

    def __init__(self):
        super().__init__()
        self.downsample = nn.Conv2d(3, 16, 3, stride=2, padding=1)
        self.upsample = nn.ConvTranspose2d(16, 16, 3, stride=2, padding=1)

    def forward(self, x):
        return self.upsample(self.downsample(x), [8, 8])

# float model output shape: [B, 16, 8, 8]
# Quantized model output shape: [B, 16, 7, 7]

Log output

No response

[github-actions]

Stale issue message