-
Notifications
You must be signed in to change notification settings - Fork 1.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Fine tune TrOCR on IAM Handwriting Database using Seq2SeqTrainer #412
Comments
Managed to get the data from the origin, now another problem, I follow your example and I get no model saved... do I do anything wrong?
|
If you provide the |
I did a:
And it saved... the old model? Anyway, I would like to save the best model. I think the save doesn’t care about the best model, so will just save every save_steps regardless of which step had the better loss. Am I wrong? Should I evaluate by epoch? |
Thank you so much! Side question: do you have any scripts/docs how to train a foreign language TrOCR model? I mean here by foreign language especially Hebrew. |
Refer to this thread: huggingface/transformers#18163 |
Sorry for disturbing, I'm a novice in BERT... untill now I worked only with Kraken OCR, still neural networks but a little different.
|
|
Finally I trained it on |
I'd recommend starting with 5 training examples and see if the model is able to overfit them |
Ok, I'll do that! Should I keep |
Trained with 16 samples:
Test output:
Even worse. |
That means there's a bug in data prepatation/hyperparameter settings/model configuration. I recommend this guide for debugging: https://karpathy.github.io/2019/04/25/recipe/ |
Seems the IAM dataset is not public anymore, any other location?
Trying to download, output:
The text was updated successfully, but these errors were encountered: