We get few users report this issue. This would be implemented shortly. We talked about two different options

1. AIBrix should provides an abstraction to provide all available models. the challenge is current plugin design does not like to manipulate the response. Technically, it should aggregate the models (including lora adapter) from the cache. data plane traffic works out-of-box in multi-model scenarios.
2. refactor plugin to provide model scope /v1/models return results. That means it only care about single base + all the related adapters. but seems this way need more efforts and can not meet production users's need.

Currently, I prefer the 1st way and we need to do such trick for such api. It brings another problem on the api compatibility and we should revisit it now. #732