[RFC]: Join the MultiLora and MultiLora Dynammic Serving feature develop #396
Description
Motivation.
We would like to join the MultiLora and MultiLora Dynammic Serving feature develop.
Proposed Change.
We want to implement the following：
1、The MultiLora usage should be the same as vllm, see https://docs.vllm.ai/en/latest/features/lora.html
2、What's more, for production environments use, the dynamically serving LoRA Adapters should be persistence. That means when the docker is restarted in some case, the load/unload lora adapters should not roll back to the initial state.
Feedback Period.
we plan to finnish this work in two weeks
CC List.
Any Other Things.
No response