fix

demo/infer.py

@@ -36,6 +36,7 @@ class LiveCCDemoInfer:
 
     @spaces.GPU
     def __init__(self, model_path: str = None, device_id: int = 0):
+        os.system('pip install flash-attn --no-build-isolation')
         self.model = Qwen2VLForConditionalGeneration.from_pretrained(
             model_path, torch_dtype="auto", 
             # device_map=f'cuda:{device_id}',