Do we need to freeze embedding layer and the lm_head as well during the Llama-pro style training ?

[hills-code]

yes, I freeze those two modules as well in the training.

[shamanez]

Thanks a lot