Azure document intelligence (OCR) is not stable? How to fix or set a random seed?

Hi everyone,

I am using this service (AI Document Intelligence) for OCR task. I used prebuilt invoice model. I find the automatic table detection feature of this service quite good. Many times it helps the results become clearer. However, yesterday I encountered a case where the service with 2 runs gave 2 different results. And the different location was in the table area that the service automatically detected.

I mean, in my specific case, on the paper there was actually no table available, only keyword fields separated by spaces. And this service automatically determined that there was a table there. However, after 2 runs, the table in this position had a different number of columns. It makes my final result has problem.

So, is there a parameter related to random seed that exists? Or is there any way to keep the result after multiple runs?

with open(pdf_path, "rb") as f:
        poller = client.begin_analyze_document(
            model_id="prebuilt-invoice",
            analyze_request=f,
            content_type="image/png"
        )

Note: I am using free tier, and it has remaining free transaction amount.