diff --git a/src/asr/fairseq_mms/local/model.py b/src/asr/fairseq_mms/local/model.py
index 12e3842..e866b5a 100644
--- a/src/asr/fairseq_mms/local/model.py
+++ b/src/asr/fairseq_mms/local/model.py
@@ -26,6 +26,7 @@ async def inference(self,  request: ModelRequest):
         wav_file = request.wav_file
         ory_sample, sr = librosa.load(wav_file, sr=16000)
         inputs = self.processor(ory_sample, sampling_rate=16_000, return_tensors="pt")
+        inputs = inputs.to(self.device)
         with torch.no_grad():
             outputs = self.model(**inputs).logits