Problem with load trained model

Part 1 (2019)
1

Hi all.
I tried many times until I dare ask a question here. Would appreciate help.
I tried to load a trained model to the local env, after I saved.
But, when I run to load, I got this issue:
AttributeError: 'Conv2d' object has no attribute 'padding_mode'

I tried updating the version, updating fast.ai, but to no avail.

After I failed locally, I tried to raise my trained weights to the same environment I had trained, but only to try and test whether it worked. And to my surprise I get the same error.

I add the code I try to load model only:

defaults.device = torch.device('cpu')
img = open_image('test-image.jpeg')
learn = load_learner('/notebooks/course-v3/nbs/dl1')
pred_class,pred_idx,outputs = learn.predict(img)
pred_class

An identical problem occurred when I tried to upload the model to render web server.

Please, any help.
Thanks.

2

This one’s hard to debug, it looks like it’s loading in a model it didn’t expect, but it’s hard to say more without seeing what you’re doing.

Could you post a minimal notebook that shows how the issue occurs? I.e. how did you save the model, and how are you loading it?

3

Thank you.
My problem solved by loading the .pth file like an Jeremy example, and now it’s looks good.

thanks.