Unexpected output shape from nnx.Conv with 'valid' padding

[Impure-King]

I'm encountering an unexpected output shape when using nnx.Conv with padding='valid'.
Here are the details:

• Input tensor shape: (32, 28, 28, 1)
• nnx.Conv layer parameters:
  • in_features=1
  • out_features=64
  • kernel_size=(4, 4)
  • strides=(2, 2)
  • padding='valid'
• Expected output shape (based on the standard convolution formula): (32, 13, 13, 64)
• Actual output shape obtained from the code: (32, 14, 14, 64)
  The spatial dimensions (height and width) are 14, while the standard formula predicts 13 for 'valid' padding with these parameters.
  I'm wondering if there's a specific behavior or setting in nnx.Conv that might explain this discrepancy. Could anyone shed light on why I might be getting an output shape of (14, 14) instead of (13, 13) in the spatial dimensions with 'valid' padding?
  Thanks in advance for any insights!

[Impure-King]

Just wanted to provide an update on my question. It's possible that I might have made a mistake in my code or in reporting the padding used.
I realized that with padding='same', the expected output shape for the given input and parameters would indeed be (32, 14, 14, 64). This matches the output I observed.
I apologize for any confusion caused and appreciate everyone's time and willingness to help! I'll mark this as resolved and close the discussion.
Thanks again!