diff --git a/modelscope/preprocessors/nlp/sentence_piece_preprocessor.py b/modelscope/preprocessors/nlp/sentence_piece_preprocessor.py
index 6b0b76e1..fbaa7ace 100644
--- a/modelscope/preprocessors/nlp/sentence_piece_preprocessor.py
+++ b/modelscope/preprocessors/nlp/sentence_piece_preprocessor.py
@@ -38,3 +38,14 @@ class SentencePiecePreprocessor(Preprocessor):
 
     def __call__(self, data: str) -> torch.Tensor:
         return torch.tensor(self.tokenizer.encode([data]), dtype=torch.long)
+
+    def decode(self, tokens, **kwargs):
+        """Decode the tokens to real text.
+
+        Args:
+            tokens: The output tokens from model's `forward` and `generate`
+
+        Returns:
+            The actual text.
+        """
+        return self.tokenizer.decode(tokens)