remove uint8 cast in weight calculation for boundary CE loss and fix ZeroDivisionError
Summary: The uint8 cast means that the floating point non_bd_weight is never assigned Reviewed By: wat3rBro Differential Revision: D30176377 fbshipit-source-id: 013602bb4313393f220ee0f1510bf1ff83bd56fc
Showing
Please register or sign in to comment