Low scores after extraction with Multi-page documents.

Hi Community.

I have a MS Syntex model that contains PDF files of more than one page, some are up to 100 pages long.

When the training of the model is completed, the scores are 99% accuracy. But when testing the model with the Quick Test or loading a document one of the documents used in the training (multiple pages) the score is 0% in all extracted fields.

It was also validated that if the same test is done with only the first page of the document the score is 99%.

Currently the configured flow depends on the Confidence Score to take certain paths. In this case it does not matter that all the fields were correctly extracted, it always goes for the negative condition because in the end the score is less than 20%.

I appreciate any recommendations you can give me. Mario Fulan spucelik