Lines 400 - 401 of openfold/model/primitives.py module have the following comment
# DISCREPANCY: c_hidden is not the per-head channel dimension, as
# stated in the supplement, but the overall channel dimension.This comment is contradictory to description of c_hidden in the docstring of the init method as well as how c_hidden is used in the output dimension of the linear projections. Can the authors please clarify why there is a dicrepancy and how c_hidden is the overall channel dimension?
Lines 400 - 401 of openfold/model/primitives.py module have the following comment
This comment is contradictory to description of c_hidden in the docstring of the init method as well as how c_hidden is used in the output dimension of the linear projections. Can the authors please clarify why there is a dicrepancy and how c_hidden is the overall channel dimension?