Task 02: Deploy the RAG flow to an online endpoint

1. Open a terminal in the root directory of your project.

2. Run the following command to build your flow and create a Docker image:

   pf flow build --source src --output dist --format docker
   

   This command packages your flow and outputs it in the dist directory in Docker format.

1. In your terminal, run the following command:

   export PYTHONPATH=./src:$PYTHONPATH
   

Skipping this step will result in a ModuleNotFoundError: No module named ‘chat_request’. Sometimes you will see errors related to azure_config also . This is because the files in /src cannot be found. To test your path, you can type echo $PYTHONPATH. You should see ./src. If you need to modify a corrupted path, you can type export PYTHONPATH=”“ to clear it and try again.

1. Open the deployment script: util/deploy_moe.py

2. Go to util/deploy_moe.py, line 77 and update it with the following code to configure your flow to work with the AI Foundry Test Chat interface.

        model=Model(
            name="ragflow",
            path=flow_path,  # path to promptflow folder
            properties=[ # this enables the chat interface in the endpoint test tab
                ["azureml.promptflow.source_flow_id", "ragflow"],
                ["azureml.promptflow.mode", "chat"],
                ["azureml.promptflow.chat_input", "question"],
                ["azureml.promptflow.chat_output", "answer"]
            ]
        ),
   

3. In util/deploy_moe.py, under environment_variables on line 111, add the following values:

   "AZURE_TENANT_ID": os.environ["AZURE_TENANT_ID"],
   "AZURE_CLIENT_ID": os.environ["AZURE_CLIENT_ID"],
   "AZURE_CLIENT_SECRET": os.environ["AZURE_CLIENT_SECRET"],
   

   

4. Save changes to deploy_moe.py.

5. Open the .env file in your project folder, replace the following values and add to the file:

   AZURE_CLIENT_ID=[YOUR_SP_CLIENT_ID]
   AZURE_CLIENT_SECRET=[YOUR_SP_CLIENT_SECRET]
   

   

6. Save changes to .env.

7. In the terminal, run the following command. Be sure to replace the placeholder XXXX in both the endpoint-name and deployment-name with a unique four-digit number of your choice.

   python util/deploy_moe.py --endpoint-name rag-XXXX-endpoint --deployment-name rag-XXXX-deployment
   

   Both the endpoint and deployment names must be unique within your Azure region. If you see an error indicating that the name is already in use, simply choose different names. Please note, the deployment process can take several minutes to complete.

   This may take around 20 minutes to deploy.

8. Once the deployment completes, you should see output similar to the following in your terminal:

   

If you receive the error “Key based authentication is not permitted on this storage account,” enable the option Allow storage account key access in the Configuration section of your storage account in the Azure portal.

1. Go to your new deployment in AI Studio and open the Test tab.

2. Ask a question like this: “How can I request a refill for my prescription at Lamna Healthcare?”

   

1. Navigate to your deployment in AI Studio and open the Consume tab.

2. Select the Python tab, then copy the example code.

   

3. Save it as a test.py file in the ./local directory within your repository.

4. Before running the program, update the test.py file with the request data and your deployment API key for accessing the deployment.

   

   Request data:

   {
     "question": "How can I request a refill for my prescription at Lamna Healthcare?",
     "chat_history": []
   }
   

5. Now, you’re ready to run the test.py program.

   python local/test.py