Train and publish your document processing model
After you create your document processing model, you can train, test, and publish it to make it available.
Train and validate your model
Select Next to check your selected form fields. If everything looks good, select Train to train your model.
When training is completed, select Go to Details page on the Training complete screen.
Quick-test your model
On the details page, select Quick test.
You can either drag a document or select Upload from my device to upload your test file. The quick test should only take a few seconds before displaying the results.
Select Start over to run another test, or Close if you're finished.
Troubleshooting tips
If you have trouble training your model, try these suggestions:
If you don't see your training document, go to Training document isn't displayed on the document processing model details page for a possible resolution.
Make sure your data meets the guidelines listed in document processing model requirements and limitations.
Learn how you can improve the performance of your document processing model.
Download sample material and use it for testing.
Publish your model
If you're happy with your model, you can select Publish to publish it. When publishing is complete, your model is promoted as Published and is ready to be used. More information: Publish your model in AI Builder
After publishing your form-processing model, you can use it in a Power Apps canvas app or in Power Automate.
Limitations
Calls made per environment across document processing models, including prebuilt models like receipt processing and invoice processing, are limited to 360 calls per 60 seconds.
If file processing exceeds 90 seconds during a quick test, you encounter a 408 - Dependency Timeout error. The reason is that the quick test is designed with a 90-second time-out limit. To ensure smooth testing, use files with fewer pages and a smaller size. For larger files, we recommend that you create a Power Automate flow to validate them. The Power Automate flow uses the
Predict
operation, which offers a 60-minute time-out. This makes it ideal for testing large files efficiently.