This message was deleted.

Hey there, it seems like we don’t have support for this usecase. What you can do is to save the processor and the FlaxWhisper separately with

bentoml.transformers.save_model

, then you probably will need to implement a custom runner to use these two. Additionally, this use case is fun to implement because we actually don’t have support for TPU just yet, so you also have to implement a custom strategies for it as well.

Ok thank you very much for the answer ! I guess I will go for the custom runner then.. I don't think I will use it on TPU just yet so that's not an issue for now