Azure Document Intelligence Model Training Error - InternalServerError

Question

I have been experiencing issues with training models in Azure Document Intelligence. In my subscription, I have both F0 and S0 tiers. Recently, I purchased a plan for 20,000 pages on the S0 tier. The models that have a "Succeeded" status were copied from the S0 tier, but as with any training process, I need to update some details. However, when I attempt to do so, I receive an error.

I found a similar case reported by another user in a forum and implemented the suggested solutions, but none have worked. I urgently need help to resolve this issue.

Thank you in advance for your support.

Answer 1

Hi Sebastian Villa,

Thanks for posting your question in Microsoft Q&A, apologize for any inconvenience caused on this.

The "InternalServerError: An unexpected error occurred" is a generic error that can be challenging to diagnose. Here are some steps and quick workarounds that might help resolve this issue:

1. Retry Model Training with a New Model ID:
   1. Instead of updating an existing model, try training a new model with a different ID.
   2. This will help determine if the issue is related to the previous model's metadata.
2. Check Data Encryption & Key Vault Permissions: If you are using Customer-Managed Keys (CMK) for data encryption, ensure that:
   1. The Azure Key Vault is accessible and not misconfigured.
   2. The AI service has proper access to decrypt stored data.
   Try switching to Microsoft-Managed Keys temporarily to see if encryption issues are causing the error.
3. Validate Training Data Formatting: Ensure that your documents meet Azure Document Intelligence format requirements:
   1. PDFs and images should be under 200 pages per document.
   2. Use diverse and correctly labeled training samples.
   3. Make sure there are no duplicate bounding boxes in labeled tables.
4. Use Fields.json from a Successful Model:
   1. If a previous model trained successfully, try copying the fields.json file from it.
   2. Modify only the necessary details before retraining the model.
5. Retry Training in a Different Region:
   1. Check the Azure Status Page for any regional issues affecting model training.
   2. If issues persist, deploy and train the model in another Azure region.

If the issue persists, collect logs using Azure CLI and provide the correlation IDs from failed model attempts. This will help the me to dig into the issue.

I hope these steps help resolve your issue. If you need further assistance, feel free to ask!

Thanks