I am wondering if I want to apply reft to one layer in my model, and use prefix + suffix (multiple positions) like "f2+l2". And I am using the function get_intervention_locations() in make_multiple_position_supervised_data_module to create the input "intervention_locations" for the reft model. What number should the num_interventions be? Cuz I just got an empty list from the code below. I am wondering if num_interventions argument for function get_intervention_locations() should be * 2? Though I only have one layer of intervention, there are actually two interventions right? One for prefix and one for suffix? Thank you very much!
@comeandcode Yes, if share_weights=False, then the num_interventions is 2 * intervening layer number. For instance, you are intervening only on the last layer, then num_interventions=2.
I am wondering if I want to apply reft to one layer in my model, and use prefix + suffix (multiple positions) like "f2+l2". And I am using the function
get_intervention_locations()
inmake_multiple_position_supervised_data_module
to create the input "intervention_locations" for the reft model. What number should thenum_interventions
be? Cuz I just got an empty list from the code below. I am wondering ifnum_interventions
argument for functionget_intervention_locations()
should be * 2? Though I only have one layer of intervention, there are actually two interventions right? One for prefix and one for suffix? Thank you very much!