Base model and vector merging

[deniaud]

Hi, can you suggest how to merge the base model and the trained vector?

[vgel]

Hmm, not sure if it's possible since the library wraps extra code around the model to add the vector. Maybe by adding the vector to the appropriate layer bias? But it might be architecture-dependant based on normalization... I've been thinking about this as well so open to suggestions here!