model_name as an additional optional parameter for azure open ai client.

[HerrIvan]

Presentation of the new feature

The Azure OpenAI documentation explains that model instances have two properties: deployment_name and model_name. While model_name refers to the well-known OpenAI models (e.g., GPT-4, GPT-3.5-turbo), deployment_name specifies a specific instance within an Azure tenant. There can be cases where these names differ, such as deploying multiple versions of the same model. If these names do not match, the code may attempt to select a tokenizer based on the deployment_name and fail. Allowing model_name to be passed as an optional parameter would enable the correct selection of the tokenizer, avoiding errors.

Where does it fit in Outlines?

Since Azure OpenAI is already integrated, fixing this costs nothing.

Are you willing to open a PR?

Yes I will, it takes like 20 ASCII characters max to fix this while being backwards compatible.

[HerrIvan]

Perfect, thanks Rémi