A model deployment makes an AI model available for applications to call. An Azure OpenAI resource needs a deployed model before it can process requests. For a bookstore, that could mean making a recommendation model available to suggest books from customer preferences. section 7.1
A practical deployment has a few moving parts:
- Create an Azure OpenAI resource, then give the deployment a name that describes its job. section 7.1
- Choose a model and version, plus a capacity setting for requests. The example uses GPT-4o version 2024-08-06, Standard capacity 1. section 7.1
- For production, track deployments and use approval gates when promoting models from development. section 7.4
C:/> az cognitiveservices account deployment create--resource-group BanningRG --name banningBooksOpenAI--deployment-name book-recommender --model-name gpt-4o--model-version 2024-08-06 --model-format OpenAI--sku-name Standard --sku-capacity 1
The Azure CLI command creates a named model deployment on an existing Azure OpenAI resource. Its model, version, and capacity are explicit rather than left to guesswork. section 7.1
Book recommendations
A customer who liked *1984* asks for similar dystopian novels. The application sends that review as a prompt to the deployed model’s endpoint, using the deployment name, API version, and API key in its request. The model can then return recommendations with brief descriptions. section 7.1
Deployment is also part of an operational workflow. Azure AI Foundry brings training, fine-tuning, deployment, and monitoring into one workspace, with version tracking, governance checks, and pipelines. Getting an endpoint working is the first step; keeping production changes controlled is the next. section 7.4
