diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/.train.py.swp b/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/.train.py.swp
index 135c410f4..560db54c3 100644
Binary files a/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/.train.py.swp and b/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/.train.py.swp differ
diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/train.py b/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/train.py
index 567ceafa7..2531e14aa 100755
--- a/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/train.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/train.py
@@ -171,6 +171,12 @@ def add_model_arguments(parser: argparse.ArgumentParser):
         default=4,
         help="How many left context can be seen in chunks when calculating attention.",
     )
+    parser.add_argument(
+        "--group-size",
+        type=int,
+        default=4,
+        help="How many left context can be seen in chunks when calculating attention.",
+    )
 
 
 def get_parser():