Use L40 instead of RTX2070; give kinematics head embedding as an additional input for the decay mode head; reduce learning rate as the batch size is now way bigger - L40 c1b0d84 LauritsT commited on May 7
