Model evaluation: Published vs Last trained version

[JZ_]

Last week I added some new documents (PDF-files) to existing collections, to improve the functionality of my Order Processing Model. After the model finished the training, I noticed the Accuracy score of my model went up from 89% to 97%.

Great news of course, but after checking the accuracy score details, I noticed the Model evaluation for the 'Last trained version' looks differently compared to the 'Published version' and for many collections/order lines 'N/A' is now shown in the Last trained version.

1. Why does the Model evaluation look differently for the 'Last trained version' compared to the 'Published version'?
2. Is there any risk my model will not work anymore if I publish the 'Last trained version'?
   
   

[plarrue]

Hi @JZ_ ,

Thanks for reaching out . Would you please share the model ID so we can investigate on our end ?

Here’s how to get the Model ID

Expand AI Builder > Models > Select your model > Select Settings

thanks

[JZ_]

@plarrue Model ID is fed7c6c1-dc22-4a43-85b9-57f49efd1d86