Azure AI Document Intelligence
An Azure service that turns documents into usable data. Previously known as Azure Form Recognizer.
Document Intelligence Training model InternalServerError: An unexpected error occurred.
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(294.40000000000003, 5%, 38%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3ED%3C/text%3E%3C/svg%3E)
Donny
0
Reputation points
Hi,
I tried 3x to train a custom extraction model on Azure Document Intelligence but everytime I got this error message:
InternalServerError: An unexpected error occurred.
Azure AI Document Intelligence
{count} votes
-
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(220.8, 32%, 31%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3ERS%3C/text%3E%3C/svg%3E)
Ravada Shivaprasad 1,115 Reputation points Microsoft External Staff Moderator2025-08-26T21:19:21.8+00:00 Hi Donny
If you're encountering the error message “InternalServerError: An unexpected error occurred” while trying to train a custom extraction model in Azure Document Intelligence, there are several potential causes worth checking. This error often stems from violations of input requirements, such as exceeding file size limits, using unsupported formats like password-protected PDFs, or having misaligned labels in your training data.
Another common issue involves encryption settings—if you're using customer-managed keys via Azure Key Vault, ensure the service has proper access and permissions. Switching temporarily to Microsoft-managed keys can help isolate encryption-related problems. Additionally, metadata conflicts may arise when updating an existing model, so it's advisable to train a new model with a fresh ID instead.
Some users have also reported regional instability, so trying a different Azure region might resolve the issue. Lastly, ensure your dataset is within the documented thresholds; training with too many documents or overly complex label files can trigger server errors.
Reference links : Azure AI Document Intelligence documentation , Improve document output quality with confidence, grounding, and in-context learning
Hope it Helps!
Thank you
-
Donny 0 Reputation points
2025-08-27T07:06:30.0933333+00:00 Hi @Ravada Shivaprasad ,
thanks for your prompt response.
Actually this is not a new training set. I have created several models but this time it just not working. I followed all the guidelines so far.
How can I get the exact error message? Currently Im tapping in the dark with this generic error message.
-
Donny 0 Reputation points
2025-08-28T13:45:07.7833333+00:00 Hi @Ravada Shivaprasad ,
can you please help me out here?
Thanks. -
Ofek Ben Eliezer 0 Reputation points2025-08-28T13:52:58.2233333+00:00 Yes, you can get a precise error. Train or re-trigger the build via REST or SDK and then read the operation status payload. The long-running build operation returns a JSON body with
error.code,error.message,error.innererror, anderror.details[target]that pinpoints the failure. Also enable Diagnostic settings to Log Analytics to capture request and response logs.How to get the exact error
Option A - REST
Start a build as usual. The response includes
Operation-Locationwith an operation id.- Poll the operation:
The JSON will includeGET https://{your-endpoint}/documentintelligence/operations/{operationId}?api-version=2024-11-30statusand, on failure, anerrorobject withinnererroranddetails[target].
Option B - SDK try/catch
Wrap your build call in a try/catch and print the service error. For example in .NET or Python, failed builds raiseHttpResponseErrorthat includes the serviceerrorpayload.Also turn on platform logs
In the Azure portal go to your Azure AI resource → Monitoring → Diagnostic settings.
Add a setting that sends Audit, RequestResponse, and AllMetrics to a Log Analytics workspace.
- Query logs. Example KQL to see recent failures:
AzureDiagnostics
| where ResourceProvider == "MICROSOFT.COGNITIVESERVICES" | where toint(StatusCode) >= 400 | project TimeGenerated, OperationName, StatusCode, ResultType, ResultDescription, ActivityId_g, CorrelationId_g | sort by TimeGenerated desc This gives you timestamps, operation names, and correlation ids you can share with support.
Common hidden causes to confirm quickly
Password-protected or encrypted PDFs, unsupported formats, oversized files, or images out of bounds.
Label mismatches or malformed label JSON.
Attempting to update a model in place that has conflicting metadata.
All of the above will surface in
error.innererrororerror.details[target]when you query the operation.If you still see only InternalServerError
Capture the
apim-request-idorx-ms-request-idfrom the failed response headers and open a support ticket with the id, timestamp, region, and operation id. This is how the service team traces backend failures.Try the same build in a second region to rule out a regional incident.
Enable retries with exponential backoff in your client for transient 5xx.
Document Intelligence uses a unified error schema. Every 4xx or 5xx returns structured JSON with
errorandinnererror, and build operations expose their final error via the operation status endpoint. Diagnostic settings give you platform-level visibility if the Studio view is too generic.If you want, I can also give you a minimal
curland a short Python snippet that print theinnererroranddetails[target]so you can paste them directly in a follow-up. - Poll the operation:
Sign in to comment
Sign in to answer