diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/.conformer.py.swp b/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/.conformer.py.swp
index f8183062f..b641a3e9a 100644
Binary files a/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/.conformer.py.swp and b/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/.conformer.py.swp differ
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 827c5aed1..9265c4a03 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/conformer.py b/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/conformer.py
index 3a3a5e43c..2c51a9574 100644
--- a/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/conformer.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless_gtrans/conformer.py
@@ -86,6 +86,7 @@ class Conformer(EncoderInterface):
         short_chunk_size: int = 25,
         num_left_chunks: int = -1,
         causal: bool = False,
+        group_size: int = 12,
     ) -> None:
         super(Conformer, self).__init__()
 
@@ -124,7 +125,7 @@ class Conformer(EncoderInterface):
         self.encoder = ConformerEncoder(encoder_layer, num_encoder_layers)
         self._init_state: List[torch.Tensor] = [torch.empty(0)]
 
-        self.group_size = 6
+        self.group_size = 
         self.alpha = nn.Parameter(torch.rand(self.group_size))
         self.sigmoid = nn.Sigmoid()
         self.layer_norm = nn.LayerNorm(512)