diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py b/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
index e55185051..75d24106a 100644
--- a/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
@@ -1795,7 +1795,6 @@ class Conv2dSubsampling(nn.Module):
         )
 
         self.convnext1 = nn.Sequential(ConvNeXt(layer2_channels),
-                                       ConvNeXt(layer2_channels),
                                        ConvNeXt(layer2_channels))
 
         self.conv2 = nn.Sequential(
@@ -1807,7 +1806,6 @@ class Conv2dSubsampling(nn.Module):
                                    ))
 
         self.convnext2 = nn.Sequential(ConvNeXt(layer3_channels),
-                                       ConvNeXt(layer3_channels),
                                        ConvNeXt(layer3_channels))