Problem

Some users of larger companies will want to use an existing LLM deployment in MS Azure AI Foundry instead of creating one through translate5 and train it.

Reason: They might not get rights in Azure for model deployment and training. And for cost reasons they might want to use an existing deployment instead of training one. And go with the RAG approach for prompting instead of training.

Solution

Instead of selecting a model from a model list for Azure it should be possible to pass a deployment name. And in this case it should be enough to provide Azure credentials with access to use that model, but not to provide new deployments or train them.

So configuration of Azure must be changed to allow the different usage scenarios and usage of Azure AI Foundry in language resource creation need to reflect that different usage scenarios as well.

But even if you have rights to allow for model deployment through translate5 you should be able to pass the name of an existing deployment on language resource creation.