Accessing the net's class source code

Hi,
i just ran several different versions of my network and i saved my models using torch.save().
i now want to reproduce the experiment with the best accuracy but i don’t remember which version of my network’s Class was used in that experiment.
when i tried to load the model with torch.load() i got the following massage (which exactly describes the situation i’m in):
“you can retrieve the original source code by accessing the object’s source attribute or set Module.dump_patches = True` and use the patch tool to revert the changes.”

problem is i don’t see any “source attribute” in my loaded object and don’t know what the patch tool is.

does enyone know i can i accsess the source code of the network’s class definition that was used when the saved model was created?

thanks alot

1 Like

anyone??
thanks again

bumping this up again.
still need help on that…

For what it’s worth, if you run a Python interactive session and do

torch.nn.Module.dump_patches = True
model=torch.load("mymodel.model")

…pytorch will save one or more *.patch files which you can read in a text editor. They read like a typical file diff, showing the differences between your current code and what was loaded. You can use the command-line tool “patch” if you want to automatically change your current source files to be the same as the the loaded code - or you can make any changes you feel are necessary manually based on the diff. Or just ignore the warning, if the diff looks like only superficial changes have been made.

2 Likes

Thanks so much!
i will try it soon and update

@docsjoo - thanks so much!
worked perfectly. exactly what i needed.

thanks