diff --git a/egs/librispeech/ASR/.prompt_tuning.sh.swp b/egs/librispeech/ASR/.prompt_tuning.sh.swp
index eeb94631f..9c9dd5016 100644
Binary files a/egs/librispeech/ASR/.prompt_tuning.sh.swp and b/egs/librispeech/ASR/.prompt_tuning.sh.swp differ
diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.prompt_tuning.py.swp b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.prompt_tuning.py.swp
index a0cce3133..fcfd1bdf4 100644
Binary files a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.prompt_tuning.py.swp and b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.prompt_tuning.py.swp differ
diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/prompt_tuning.py b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/prompt_tuning.py
index 758978e03..8ec4a5462 100755
--- a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/prompt_tuning.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/prompt_tuning.py
@@ -1591,7 +1591,7 @@ def run_adapter(rank, world_size, args, wb=None):
     for n, p in model.named_parameters():
         p.requires_grad = False
     
-    prompt = torch.nn.Parameter(torch.randn(50, 512))
+    prompt = torch.nn.Parameter(torch.randn(50, 512)).to(device)
     
     '''
     optimizer_adapter = ScaledAdam(