Using learn.tta on kaggle

I am trying to use .tta
The first epoch runs well.
Then it stops with ‘file/directory not found’ error.
It appears .tta creates a ‘models’ folder under ‘working’ (on kaggle).
Then it tries to find ‘model.pth’ under its newly created ‘models’ folder.
Attached the first successful tta epoch and lateron error message.
Thank you in advance.

Hey, would you mind sharing that notebook? Its easier for others to understand what you did and recreate the error :slight_smile: A random guess from a quick glance at the error message would be that your learner is using a callback that tries to load/save the model (like early stopping) which you could just remove for inference.

1 Like

Thank you!

As you hinted, removing ‘savemodelcallback’ saved it.
My learner definiton was in another notebook, inference in another notebook.
So I share my learner definition and the removal step. (was also more efficient then sharing my untidy notebook with lots of comments in turkish : -)

1 Like