hotfix for decoding logits issue

app.py

@@ -32,7 +32,7 @@ def predict_and_decode(input_file):
   speech = load_data(input_file)
   #tokenize
   input_values = processor(speech, return_tensors="pt", sampling_rate=16000).input_values
-  logits = model(input_values).logits
+  logits = model(input_values).logits.cpu().detach().numpy()[0]
   vocab_list = list(processor.tokenizer.get_vocab().keys())
 #   #Take argmax
 #   predicted_ids = torch.argmax(logits, dim=-1)