diff --git a/egs/librispeech/ASR/incremental_transf/.identity_train.py.swp b/egs/librispeech/ASR/incremental_transf/.identity_train.py.swp
index 0d0ea3fc4..2dde7af4a 100644
Binary files a/egs/librispeech/ASR/incremental_transf/.identity_train.py.swp and b/egs/librispeech/ASR/incremental_transf/.identity_train.py.swp differ
diff --git a/egs/librispeech/ASR/incremental_transf/identity_train.py b/egs/librispeech/ASR/incremental_transf/identity_train.py
index 56f93c676..e5bc815c8 100755
--- a/egs/librispeech/ASR/incremental_transf/identity_train.py
+++ b/egs/librispeech/ASR/incremental_transf/identity_train.py
@@ -979,10 +979,9 @@ def run(rank, world_size, args):
     except: pre_trained_model = torch.load('/home/work/workspace/icefall/egs/librispeech/ASR/incremental_transf/conformer_12layers.pt')
     pre_trained_model = pre_trained_model['model']
     transducer_model.load_state_dict(pre_trained_model, strict=True)
+    transducer_model.to(device)
     model = get_interformer_model(transducer_model.encoder, params)
 
-    del pre_trained_model
-
     for n, p in model.named_parameters():
         if 'pt_encoder' in n:
             p.requires_grad = False
diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.checkpoint.py.swp b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.checkpoint.py.swp
index b7aa68526..615e63fa8 100644
Binary files a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.checkpoint.py.swp and b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.checkpoint.py.swp differ