Check how good the latest checkpoint is and decide if we want to keep it. There are cases where CL breaks the model (see image below) in which case we want to continue using the previous one.
Drift is detected in the gray vertical dashed line and after CL the model predicts basically 0 values:
Check how good the latest checkpoint is and decide if we want to keep it. There are cases where CL breaks the model (see image below) in which case we want to continue using the previous one.
Drift is detected in the gray vertical dashed line and after CL the model predicts basically 0 values:
