Semantics of passing tuple for stride in nn.Conv1D

1

The documentation for the Conv1D layer states that a tuple or int can be passed as stride.

Does this tuple have to be one-dimensional? If not, what are the semantics?

I tried to dig into the code to find out, and it seems multi-dimensional tuples are passed along to the underlying C++/CUDNN code. I did not in understanding it though.

2

in the case of Conv1d, indeed a single int or a 1-element tuple (example: (3, ) ) can be passed

3

Thank you. I will submit a PR to clarify it in the docs.