diff --git a/modelscope/models/nlp/palm_v2/modeling_palm.py b/modelscope/models/nlp/palm_v2/modeling_palm.py
index c2121cfd..127b5440 100644
--- a/modelscope/models/nlp/palm_v2/modeling_palm.py
+++ b/modelscope/models/nlp/palm_v2/modeling_palm.py
@@ -1170,7 +1170,6 @@ class Translator(nn.Module):
         results['batch'] = batch
 
         for step in range(max_length):
-            self.logger.info(f'step: {step + 1} / {max_length}')
             decoder_input = alive_seq[:, -1].view(1, -1)
 
             # Decoder forward.
diff --git a/modelscope/models/nlp/palm_v2/palm_for_text_generation.py b/modelscope/models/nlp/palm_v2/palm_for_text_generation.py
index 7f8e918b..e432cc58 100644
--- a/modelscope/models/nlp/palm_v2/palm_for_text_generation.py
+++ b/modelscope/models/nlp/palm_v2/palm_for_text_generation.py
@@ -32,9 +32,9 @@ class PalmForTextGeneration(TorchModel):
         replace_tokens_bert = (('[unused0]', ''), ('[PAD]', ''),
                                ('[unused1]', ''), (r' +', ' '), ('[SEP]', ''),
                                ('[unused2]', ''), ('[CLS]', ''), ('[UNK]', ''))
-        replace_tokens_roberta = ((r' +', ' '), ('<mask>', '<q>'), ('<pad>',
-                                                                    ''),
-                                  ('<s>', ''), ('</s>', ''), ('<unk>', ' '))
+        replace_tokens_roberta = ((r' +', ' '), ('<mask>', '. '),
+                                  ('<pad>', ''), ('<s>', ''), ('</s>', ''),
+                                  ('<unk>', ' '), ('<q>', '. '))
 
         strings = [self.tokenizer.decode(pred_ids) for pred_ids in ids_list]
         for _old, _new in replace_tokens_bert: