Skip to content

README.md

Latest commit

 

History

History
164 lines (106 loc) · 6.15 KB

README.md

File metadata and controls

164 lines (106 loc) · 6.15 KB

1-click deployment

The new way to deploy this solutions is using the 1-click deployment. This is the recommended way to deploy this solution.

Pre-requisites

  • You will need to have an Azure subscription and be able to create resources in it. At least a resource group is required, and the user must have Owner permissions on it.

Deployment steps (Azure Portal)

  1. Click the "Deploy to Azure" button

    Deploy to Azure

  2. Fill in parameters

    There are no special parameters for this deployment. Optional parameters are available to customize the deployment (see below). Typically, only openAIName and openAIRGName are used to reuse an existing Azure OpenAI resource.

    Average deployment time is 10 minutes when no existing container registry is set.

Customization

The following parameters are available for customization:

  • openAIName and openAIRGName: Name and resource group for the Azure OpenAI resource to reuse, instead of creating a new one. registry will be created and images will be built pushed to it via cloning the GitHub repository.
  • namePrefix: Prefix for all resources created by the deployment. Default is dev.
  • newOpenAILocation: Location for the new Azure OpenAI resource. Default is empty and will be ignored unless openAIName is blank.
  • skipImagesBuild: True not to build images from the remote repository. Default is False. If set to True, the deployment will not create a container registry and you will need to build the images manually (see below).
  • useWebApps: True to provision Azure WebApps instead of Container App to host images. Default is False.

Deployment steps (local)

  1. Clone the GitHub repository

    git clone https://github.com/Azure-Samples/multimodal-rag-code-execution

cd multimodal-rag-code-execution

  2. Run the deployment script

    You could run the below commands in either a Powershell or a Git Bash.

    az login

az upgrade

az bicep upgrade

az group create --name multimodal-rag-code-execution --location <location>

## [OPTION 1] USE AN EXISTING OPENAI RESOURCE
## Existing OpenAI Resource must have one model name 'gpt-4o' and another named 'text-embedding-3-large'. Please provide the name of the OAI resource and the RG name where that resource is
az deployment group create --resource-group multimodal-rag-code-execution --template-file deployment/infra-as-code-public/bicep/main-1click.bicep --parameters aiSearchRegion=eastus openAIName=<OAI_NAME> openAIRGName=<OAI_RG_NAME>

## [OPTION 2] CREATE A NEW OPENAI RESOURCE
az deployment group create --resource-group multimodal-rag-code-execution --template-file deployment/infra-as-code-public/bicep/main-1click.bicep --parameters aiSearchRegion=eastus newOpenAILocation=eastus

Screenshots of the Deployment

In the Azure portal, go to Deployments:



Check all resources being deployed:



Troubleshooting

If you see the below screenshot, please check your API web app, check its Environment Variables, and try to restart it.

Variables to check are:

  • DOCKER_REGISTRY_SERVER_URL
  • DOCKER_REGISTRY_SERVER_USERNAME
  • DOCKER_REGISTRY_SERVER_PASSWORD

If problem persists, go to WebApp Deployment Center, and make sure:

  • The container registry is connected via Admin credentials
  • The "Continuous Deployment" option is enabled


Updating web app images

There are two ways to update the web app images:

  • Option 1: Rebuilding from remote git repository
  • Option 2: Rebuilding from local

Option 1: Rebuilding from remote git repository

This is the recommended approach if you simply need to get the latest features and bug fixes.

  1. Go to the Azure Portal and open a Cloud Shell (Bash)
  2. Copy the update_webapps_from_github.sh script to the Cloud Shell
  3. Run chmod +x update_webapps_from_github.sh
  4. Run ./update_webapps_from_github.sh <resource-group-name>

Please note images will be built and pushed to the container registry sequentially. This process will take about 10 minutes. Additionally, the script will restart the API WebApp to use the new images.

Option 2: Rebuilding from local

This is the recommended way to update the images if you have made changes to the code, or if you chose not to build from the remote repository.

  1. Clone the GitHub repository

    git clone https://github.com/Azure-Samples/multimodal-rag-code-execution

cd multimodal-rag-code-execution

  2. Make all required changes to the code

  3. Run the publish-aca.ps1 script using PowerShell (or publish-webapp.ps1 in case of WebApps)

    # Make sure to run from root of the repository
.\deployment\publish-aca.ps1 -RG <resource-group-name>

    NOTES

    • script assumes you have the Azure CLI installed and logged in to the correct subscription
    • target resorce group must have been created by the deployment script
    • images will be built and pushed to the container registry sequentially. This process will take about 10 minutes.
    • All Container/Web Apps will restart to use the new images

Local development

If you want to develop locally using resources you deployed, you must first create a Service Principal and assign it Contributor permissions on the resource group and AML Workspace. Otherwise the API won't have access to the AML Workspace.

Then, configure the following environment variables in a .env file:

AML_PASSWORD=
AML_SERVICE_PRINCIPAL_ID=
AML_TENANT_ID=

You can use the set-sp-secret.sh script to create the SP and a secret with the right permissions (can be run in Cloud Shell):

./set-sp-secret.sh <sp-name> <resource-group-name>