Detecting dynamic flow?

[eliegoudout]

Hi,

I came across the notion of static/dynamic flow on PyTorch's doc. I realize that dynamic flow (that is when module calls may not be consistent between different inputs) poses an obvious problem for torchinfo. Indeed, the summary is computed through a forward pass of an input (random or full zeros or else, I've not looked at your code to figure it out), but another input might yield a different module execution.

As such, I think it would be wise to consider issuing a warning or raising an error when detecting a dynamic flow? Otherwise, the output may be misleading.

I chose the "Feature Request" tag, but it might also e considered a "Bug Report" since it's about an elementary vulnerability.

Cheers!