fix bug

app.py

@@ -212,7 +212,9 @@ bicodec_tokenizer_path = args.bicodec_tokenizer_path or os.path.join(
 gpa_model_path = args.gpa_model_path or model_base_path
 
 # Instantiate Model
-print(f"Initializing GPA Inference System on {args.device}...")
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+print(f"Initializing GPA Inference System on {device}...")
 print(f"Tokenizer path: {tokenizer_path}")
 print(f"Text tokenizer path: {text_tokenizer_path}")
 print(f"BiCodec tokenizer path: {bicodec_tokenizer_path}")
@@ -225,7 +227,7 @@ inference = GPAInference(
     bicodec_tokenizer_path=bicodec_tokenizer_path,
     gpa_model_path=gpa_model_path,
     output_dir=None,  # Will use temporary directory
-    device="cuda" if torch.cuda.is_available() else "cpu",
+    device=device,
 )
 
 # Launch Gradio Demo