Commit c380000e authored by Mostofa Patwary's avatar Mostofa Patwary
Browse files

fixed exit interval

parent 097fa65b
Loading
Loading
Loading
Loading
+3 −3
Original line number Diff line number Diff line
@@ -884,9 +884,9 @@ def train(forward_step_func, model, optimizer, lr_scheduler,

        # Exiting based on iterations        
        if args.exit_interval and iteration % args.exit_interval == 0:
            #if not saved_checkpoint:
            #    save_checkpoint_and_time(iteration, model, optimizer,
            #                             lr_scheduler)
            if not saved_checkpoint:
                save_checkpoint_and_time(iteration, model, optimizer,
                                         lr_scheduler)
            torch.distributed.barrier()
            print_datetime('exiting program at iteration {}'.format(iteration))                
            sys.exit()