The paper mentions the ability to train multiple prefixes in a single batch.

[phalexo]

Any of the code in the repo implements this?

Thanks.