[Feature]: Support Multiple Tasks Per Model

[FurtherAI]

🚀 The feature, motivation and pitch

Requesting this for V1 #11862

The idea is pretty simple, it would be nice to be able to, e.g., get generations and embeddings out of a single model. An example use case is when you have a LoRA for generation and a LoRA for embedding on top of the same base model. Deploying two vLLM servers is really inefficient for accomplishing this.

Alternatives

A lesser feature would be one task per LoRA, but it's better to be general if possible.

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.