[wenet]fix the step_xx.yaml grad_norm issue (#2288)

correct: 
       grad_norm: 28.40692710876465
error: 
       grad_norm: !!python/object/apply:torch._utils._rebuild_tensor_v2...

wenet/utils/train_utils.py

@@ -546,6 +546,7 @@ def update_parameter_and_lr(model, optimizer, scheduler, scaler, info_dict):
                 optimizer.step()
         optimizer.zero_grad()
         scheduler.step()
+        grad_norm = grad_norm.item()
 
     info_dict["lr"] = optimizer.param_groups[0]['lr']
     info_dict["grad_norm"] = grad_norm