Update app.py

app.py

@@ -66,6 +66,7 @@ def inference(audio):
           return_tensors="pt"
     ).input_values
     # decode ctc output
+    logits = model(input_values).logits[0]
     pred_ids = torch.argmax(logits, dim=-1)
     greedy_search_output = processor.decode(pred_ids)
     beam_search_output = ngram_lm_model.decode(logits.cpu().detach().numpy(), beam_width=500)