Dynamic Engine Argument Configuration for vLLM Runtime in WebUI

[hoyajigi]

Main idea

Hello, I am a user who loves and uses backend.ai well.

�In the Model Serving, there is vLLM Runtime Variant, but there is no function to dynamically specify engine arg (such as max_num_batched_tokens) and it must be embedded in the Docker image. Since this is a strongly parametric value and needs to be optimized, I hope you create a way to run the model by specifying those in the UI.

Alternative ideas

No response

Anything else?

No response