Automated Azure Deployment

[ghost-pep]

It would be great to have the needed models for the project deployed automatically using either ARM templates or bicep. This would save from long setup documentation and also ensure correctness for the project.

[michael-montana]

Hi @ghost-pep,

Thanks for raising this issue!

I could not find a reference to the specific Azure resources needed in the repository. Could anybody help clarify which resources (e.g., Azure Functions, Cognitive Services, or others) you’d like to have automatically deployed?

Based on the project structure, I assume the following might be relevant:

• Azure OpenAI Service or Cognitive Services for AI models,
• Azure Storage and Azure App Service/Functions for hosting,
• and potentially Key Vault for secrets management.

Kind regards

[ghost-pep]

Hey! These are great questions! For the code that is currently in the repo, we deployed an Azure OpenAI (AOAI) instance and then deployed models like GPT3/4 for a backing implementation for our text completion using API keys. The right way to do this would be to have a fully automated deployment that does not use keys and rather deploys our code to Azure App Service and then links to the Azure OpenAI instance using managed identities. This can all be automated using Bicep (my preferred solution over ARM templates directly).

For those looking to work on this issue, I'd start by getting a Bicep file that can deploy Azure OpenAI with models working. I would say that the overachieving way would be to fully automate the app deployment as well but that would be a massive bonus to the MVP of just getting automated AOAI up and running