Update infer.py

infer.py

@@ -50,10 +50,10 @@ def load_model(model_name, audio_tokenizer_path):
         use_flash_attention_2=True,
         use_cache=True,
     )
-    model = model.to("cuda")
+    # model = model.to("cuda")
     tokenizer = AutoTokenizer.from_pretrained(model_name)
     tokenizer_voila = VoilaTokenizer(model_path=audio_tokenizer_path, device="cpu")
-    tokenizer_voila.to("cuda")
+    # tokenizer_voila.to("cuda")
     return model, tokenizer, tokenizer_voila, model_type
 
 def is_audio_output_task(task_type):