HTTP server should vary endpoints by available models #1313
Description
Feature request
In a distributed setup we have an ingress node like this: dynamo-run in=http out=dyn.
That currently exposes endpoints to list models (that should stay always), do completions, chat completions, soon embeddings, and more in the future.
However we might only have a chat model, in which case completions and embeddings should not be advertised, and so on.
We know what kind of model we are serving because they register themselves with a ModelType.
Describe the problem you're encountering
/
Describe alternatives you've tried
No response