diff --git a/src/llamafactory/train/sft/workflow.py b/src/llamafactory/train/sft/workflow.py
index 502ac2f7..47e2bb3b 100644
--- a/src/llamafactory/train/sft/workflow.py
+++ b/src/llamafactory/train/sft/workflow.py
@@ -119,10 +119,21 @@ def run_sft(
         print("********************************************************")
         print("********************************************************")
         print("********************************************************")
-        
-        
+        print("predict_results.metrics")
         print(predict_results.metrics)
         print("********************************************************")
+        
+        
+        num_train_tokens = trainer.num_tokens(trainer.get_test_dataloader(dataset_module["eval_dataset"]))
+        print("num_train_tokens")
+        print(num_train_tokens)
+        print("********************************************************")
+        predict_results.metrics["num_train_tokens"] = num_train_tokens
+        print("new predict_results.metrics")
+        print(predict_results.metrics)
+        print("********************************************************")
+        
+        
         time.sleep(100)
         if training_args.predict_with_generate:  # predict_loss will be wrong if predict_with_generate is enabled
             predict_results.metrics.pop("predict_loss", None)