Feature request: Reconsider adding the models endpoint to OpenAI Compatibility

[marcjwojtowicz]

In #119 the models endpoint for OpenAI Compatibility was deemed unnecessary since Paddler only hosts a single model at a time.

However, there are other reasons for supporting this endpoint. For example, some tools which use OpenAI Compatibility rely on the models endpoint to retrieve available models, even if there is only one. Olla is an example of this. Other tools like Open WebUI have work-arounds to manually specify models in cases where the models endpoint is not supported. A common use case for tools calling the models endpoint is to populate dropdown lists with available models, sometimes aggregating results from multiple OpenAI Compatible endpoints.

My suggested implementation of the models endpoint for Paddler is to return a list with the single model Paddler is using while a model is loaded in Paddler, or to return an empty list of models while no model is loaded in Paddler.

All that said, I know there are several applications with OpenAI Compatibility which do not support the models endpoint, so consider this a "nice to have" feature request.

[malzag]

@marcjwojtowicz thank you for the rationale, we will do it :)