Rearrange the Pipeline to allow for multiple peft adapters on the same transformer, mostly by using named adapters for conparse and sentiment. (POS, depparse, and NER would require their own names if we use peft for those)
At training time, the easiest solution is to use a separate transformer so that the optimizer is not cluttered up with peft adapters for other annotators, such as if POS is used to retag depparse or conparse
Rearrange the Pipeline to allow for multiple peft adapters on the same transformer, mostly by using named adapters for conparse and sentiment. (POS, depparse, and NER would require their own names if we use peft for those)
At training time, the easiest solution is to use a separate transformer so that the optimizer is not cluttered up with peft adapters for other annotators, such as if POS is used to retag depparse or conparse