Deploying Azure OpenAI resource and models

Page content

• 1. Create and deploy an Azure OpenAI Service resource
• 2. Retrieve information about the resource
  • 2.1 Get the Azure OpenAI endpoint URL
  • 2.2 Get the Azure OpenAI API key
• 3. Create Azure Open AI chat and embedding models
  • 3.1. Create Azure Open AI embedding model
  • 3.2. Create Azure Open AI chat model
• 4. (Optional) Try the chat model in Azure OpenAI Studio playground

1. Create and deploy an Azure OpenAI Service resource

We’ll begin by creating an Azure OpenAI resource within the new resource group you created in the Prerequisites section. Please execute the following command to create the resource:

 
az cognitiveservices account create \
  --name "${resource_group_name}" \
  --resource-group "${resource_group_name}" \
  --location "${region}" \
  --kind OpenAI \
  --sku s0 \
  --custom-domain "${resource_group_name}" \
  --yes

To ensure the Azure OpenAI Service resource was created successfully, navigate to the Azure portal and access the resource group. Check that the Azure OpenAI resource named aoai-rag-oyd is present.

domain=$(az rest --method get --url https://graph.microsoft.com/v1.0/domains --query 'value[?isDefault].id' -o tsv)
subscription_id=$(az account show | jq -r .id)
url="https://ms.portal.azure.com/#@${domain}/resource/subscriptions/${subscription_id}/resourceGroups/${resource_group_name}/overview"

# URL to the Azure resource group to see the created resources in it.
echo "${url}"

2. Retrieve information about the resource

After creating the Azure OpenAI Service resource, you can use various commands to gather useful details about your instance. Below are examples of commands to retrieve the REST API endpoint base URL and the access keys for your new resource.

To store environment variables for later use, create a “.env” file in the root directory of your Git repository. This file will be used to securely collect and manage sensitive data like API keys, endpoints, and other configuration details.

touch .env

To set up Azure AI Search and use these credentials in the demo application, you need to retrieve the Endpoint and Key from your Azure OpenAI resource. Follow the steps below to fetch this information.

Open the Azure Open AI resource and:

• Nagivate to Resource Management
• Click Keys and Endpoint

2.1 Get the Azure OpenAI endpoint URL

You should see the Endpoint value. Copy the value and paste it into the .env file in the project root directory. It should look as the following:

After retrieving the Endpoint value from your Azure OpenAI resource, copy the value and paste it into the .env file located in the root directory of your project. Your .env file should look as following:

Command:

cat .env

Example output:

AZURE_OPENAI_ENDPOINT=<endpoint-url>

AZURE_OPENAI_ENDPOINT=$(az cognitiveservices account show \
  --name "${resource_group_name}" \
  --resource-group  "${resource_group_name}" \
  | jq -r .properties.endpoint)

# write the Azure OpenAI endpoint URL to config file
echo "AZURE_OPENAI_ENDPOINT=${AZURE_OPENAI_ENDPOINT}" >> .env

2.2 Get the Azure OpenAI API key

On the same Keys and Endpoint setting page, you should see the KEY 1 value. Copy the value and paste it into the .env file in the project root directory. Your .env file should look as the following:

Command:

cat .env

Example output:

AZURE_OPENAI_ENDPOINT=<endpoint-url>
AZURE_OPENAI_KEY=<key-1-value>

AZURE_OPENAI_KEY=$(az cognitiveservices account keys list \
	--name "${resource_group_name}" \
    --resource-group "${resource_group_name}" \
    | jq -r .key1)

# write the Azure OpenAI key to config file
echo "AZURE_OPENAI_KEY=${AZURE_OPENAI_KEY}" >> .env

3. Create Azure Open AI chat and embedding models

To set up Azure AI Search and the demo application, you need to create two Azure OpenAI models:

1. Embedding Model: Converts plain text into vector embeddings, which are essential for Azure AI Search.
2. Chat Model: Powers the demo application, allowing end-users to interact with PDF documents through chat.

3.1. Create Azure Open AI embedding model

To create an embedding model, run the comand below.

az cognitiveservices account deployment create \
	--name "${resource_group_name}" \
	--resource-group  "${resource_group_name}" \
	--deployment-name "${model_deployment_name_embedding}" \
	--model-name "${model_name_embedding}" \
	--model-version "${model_version_embedding}" \
	--model-format OpenAI \
	--sku-capacity "30" \
	--sku-name "Standard"

Let’s validate that the embedding model was created successfully. Open the Azure OpenAI Studio in the left navigation bar, under Shared resources, click Deployments.

id=$(az cognitiveservices account deployment show \
	--name "${resource_group_name}" \
	--resource-group "${resource_group_name}" \
	--deployment-name ${model_deployment_name_embedding} \
	| jq -r .id)
url="https://oai.azure.com/resource/deployments?wsid=${id}"

# URL to the Azure OpenAI Studio deployments view.
echo "${url}"

We need the embedding model and its deployment name to enable Azure AI Search to access it during the indexing process. Additionally, the demo application requires the deployment name as part of its configuration to set up Azure OpenAI. Use the following command to add these details to the .env file.

echo "AZURE_OPENAI_CHATGPT_EMBEDDING_DEPLOYMENT=${model_deployment_name_embedding}" >> .env
echo "AZURE_OPENAI_CHATGPT_EMBEDDING_MODEL_NAME=${model_name_embedding}" >> .env

Your .env file should look as the following:

Command:

cat .env

Example output:

AZURE_OPENAI_ENDPOINT=<endpoint-url>
AZURE_OPENAI_KEY=<key-1-value>
AZURE_OPENAI_CHATGPT_EMBEDDING_DEPLOYMENT=aoai-rag-oyd-embedding
AZURE_OPENAI_CHATGPT_EMBEDDING_MODEL_NAME=<embedding-model-name>

3.2. Create Azure Open AI chat model

To create a chat model, run the comand below.

az cognitiveservices account deployment create \
	--name "${resource_group_name}" \
	--resource-group  "${resource_group_name}" \
	--deployment-name "${model_deployment_name_chat}" \
	--model-name "${model_name_chat}" \
	--model-version "${model_version_chat}" \
	--model-format OpenAI \
	--sku-capacity "100" \
	--sku-name "GlobalStandard"

Let’s validate that the chat model was created successfully. Under the same Deployments view, click refresh and validate that there chat model is present.

From the chat model, we need the deployment name and api version for later use for deo application to be able to connect to the chat model. Run the following command to add those to the .env file.

echo "AZURE_OPENAI_CHATGPT_DEPLOYMENT=${model_deployment_name_chat}" >> .env
echo "AZURE_OPENAI_API_VERSION=${model_version_chat}" >> .env

Your .env file should look as the following:

Command:

cat .env

Example output:

AZURE_OPENAI_ENDPOINT=<endpoint-url>
AZURE_OPENAI_KEY=<key-1-value>
AZURE_OPENAI_CHATGPT_EMBEDDING_DEPLOYMENT=aoai-rag-oyd-embedding
AZURE_OPENAI_CHATGPT_EMBEDDING_MODEL_NAME=<embedding-model-name>
AZURE_OPENAI_CHATGPT_DEPLOYMENT=aoai-rag-oyd-chat
AZURE_OPENAI_API_VERSION=<chat-model-api-version>

4. (Optional) Try the chat model in Azure OpenAI Studio playground

Now that the chat model has been successfully deployed, you can test it in the playground! In the left navigation menu, navigate to Playground and select Chat. A chat interface with a free-form text input box will appear. Enter your query in the text box and press Enter to interact with the model.

Congratulations! You have successfully deployed the Azure OpenAI resource along with the embedding and chat models! Next, we will move on to deploying and configuring Azure AI Search.

« Document Data Management: Workshop Overview | Document Data Management: Azure AI Search »