deuch/simple-api-aoai-formrecog-sample

simple-api-aoai-formrecog-sample

Prepare AML Compute instance

• Clone the repo in your home directory on AML compute instance
• Run : !pip install -r requirements.txt
• Run : az upgrade
• Run : az login

Modify values in app.py

STORAGE_ACCOUNT_CONNECTION_STRING = "XXXXXXXXXXXXXXXXXXXXX"
container_name_documents = 'XXXXXXXXXXXXXXXX' #raw pdf files
OPENAI_API_KEY = "XXXXXXXXXXXXXXXXXXXXXX" # SET YOUR OWN API KEY HERE
AZURE_OPENAI_RESOURCE_ENDPOINT = "https://XXXXXXXXXX.openai.azure.com/" # SET A LINK TO YOUR RESOURCE ENDPOINT
FORM_RECOGNIZER_ENDPOINT = "https://XXXXXXXXXXX.cognitiveservices.azure.com/"
FORM_RECOGNIZER_KEY = "XXXXXXXXXXXXXXXXXXXXXX"
PROMPT= [XXXXXXXXXXXXX]

Deploy Azure App Services

• In the directory where app.py and requirements.txt reside, run :

  • Replace myappname, mylocation and mysub bu the desired values

  az webapp up --runtime PYTHON:3.9 --sku B1 --name myappname --location mylocation --subscription mysub

It will create and deploy an App Services Instance with the python code and the associated requirements. You can use the same command to update the app service in you make some modifications in your code

Testing

You can test with POSTMAN or curl command. You need to use a "file" field to upload the document.
Replace myfile.pdf and myapp with the desired values.

For curl with the App Service Deployment :
curl -k -X POST -F [email protected] https://myapp.azurewebsites.net/upload

Or to test locally in the AML Compute Instance :
curl -k -X POST -F [email protected] http://localhost:5000/upload

Overwrite is enabled for the blob, so you can upload a file with the same name and it will be overwritten.

If you have some HTTP 413 errors (Entity too large), you can increase this variable : app.config['MAX_CONTENT_LENGTH'] = 327680000

Best practices

• It's better and safer to use Managed Identities instead of key or access keys