diff --git a/egs/librispeech/ASR/.run_adapter.sh.swp b/egs/librispeech/ASR/.run_adapter.sh.swp
index b83ad4d92..06138da61 100644
Binary files a/egs/librispeech/ASR/.run_adapter.sh.swp and b/egs/librispeech/ASR/.run_adapter.sh.swp differ
diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.data2vec_audio.py.swp b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.data2vec_audio.py.swp
index 5c0e5dbe4..62292f92a 100644
Binary files a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.data2vec_audio.py.swp and b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.data2vec_audio.py.swp differ
diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/data2vec_audio.py b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/data2vec_audio.py
index dd0507e05..aba350f42 100644
--- a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/data2vec_audio.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/data2vec_audio.py
@@ -43,7 +43,7 @@ class TransformerEncoderAdapter(TransformerEncoder):
 
         for p in self.adapters.parameters():
         #    p.data = nn.Parameter(torch.zeros(p.size()).to('cuda'))
-            p.data = nn.Parameter(torch.randn(p.size()).to('cuda')/100.)
+            p.data = nn.Parameter(torch.randn(p.size()).to('cuda')/20.)
 
     def forward(self, x, padding_mask=None, layer=None, tgt_layer=None):
         x, layer_results = self.extract_features_with_adapter(