Prerequisites

Prompt flow SDK:

Install promptflow sdk and other dependencies:

pip install -r requirements.txt

Note: when using the Prompt flow SDK, it may be useful to also install the Prompt flow for VS Code extension (if using VS Code).

Azure AI/ML Studio:

Start an compute session. Required packages will automatically be installed from the requirements.txt file.

Setup connections

To use the llm tool, you must have an Azure OpenAI Service Resource. Create one if necessary. From your Azure OpenAI Service Resource, obtain its api_key and endpoint.

Create a connection to your Azure OpenAI Service Resource. The connection uses the AzureOpenAIConnection schema:

Prompt flow SDK:

# Override keys with --set to avoid yaml file changes
pf connection create -f ./connections/azure_openai.yml --set api_key=<resource_api_key> api_base=<resource_api_base> name=azure_openai

Ensure you have created the azure_openai connection:

pf connection show -n azure_openai

Azure AI/ML Studio:

To use the promptflow-azure-ai-language package, you must have an Azure AI Language Resource. Create a Language Resource if necessary. From your Language Resource, obtain its api_key and endpoint.

Create a connection to your Language Resource. The connection uses the CustomConnection schema:

Prompt flow SDK:

# Override keys with --set to avoid yaml file changes
pf connection create -f ./connections/azure_ai_language.yml --set secrets.api_key=<resource_api_key> configs.endpoint=<resource_endpoint> name=azure_ai_language

Ensure you have created the azure_ai_language connection:

pf connection show -n azure_ai_language

Azure AI/ML Studio:

If using Azure AI Studio, you will need to add two additional custom keys to the connection. Follow these instructions when creating a CustomConnection.

Note: if you already have an Azure OpenAI or Azure AI Language connection, you do not need to create additional connections and may substitute them in.

To use the CLU tool within Azure AI Language, you must have a deployed CLU model within your Language Resource. See this documentation for more information on how to train/deploy a CLU model. You may import the included MediaPlayer.json file to create a new CLU project. After training and deploying a model, note your project and deployment names.

Run flow

First, indicate a model deployment for the llm node in its deployment_name parameter. This must be a pre-existing deployment within your Azure OpenAI Service Resource. Consider changing other parameters, such as the llm node's temperature and max_tokens.

Now, update the CLU tool's project name (if you did not use the sample .json file) and deployment name parameters based on your CLU model.

Prompt flow SDK:

# Test with default input values in flow.dag.yaml:
pf flow test --flow .

Azure AI/ML Studio:

Run flow.

Flow description

The flow uses a LLM node to break down compound user queries into simple user queries. For example, "Play some blues rock and turn up the volume" will be broken down to "["Play some blues rock", "Turn Up the volume"]". This is then passed into the CLU tool to recognize intents and entities in each of the utterances.

This flow showcases the capabilities of CLU and a simple way to quickly test them on a deployed CLU model. Consider extending this flow to create a media app that acts upon user conversational requests, such as modifying the volume of a speaker, etc.

Contact

Please reach out to Azure AI Language (taincidents@microsoft.com) with any issues.