These are already exposed on KerasCV, and I think it is time to also expose these in KerasNLP. This will give us a class to document common model functionality to all backbones such as enable_lora and token_embedding on keras.io.
It can also open up a path for writing a custom architecture outside the library itself.
Note that I am not exposing GenerativeTask yet, as I think we should wait for https://github.com/keras-team/keras-nlp/pull/1425 there (and expose this as two separate classes CausalLM and Seq2SeqLM).
These are already exposed on KerasCV, and I think it is time to also expose these in KerasNLP. This will give us a class to document common model functionality to all backbones such as
enable_lora
andtoken_embedding
on keras.io.It can also open up a path for writing a custom architecture outside the library itself.