fix flashattn

app.py

@@ -1,5 +1,3 @@
-import os
-os.system('pip install flash-attn --no-build-isolation')
 import gradio as gr
 
 from demo.infer import LiveCCDemoInfer

demo/infer.py

@@ -32,6 +32,12 @@ class LiveCCDemoInfer:
     streaming_time_interval = streaming_fps_frames / fps
     frame_time_interval = 1 / fps
     def __init__(self, model_path: str = None, device_id: int = 0):
+        self.model = Qwen2VLForConditionalGeneration.from_pretrained(
+            model_path, torch_dtype="auto", 
+            device_map=f'cuda:{device_id}', 
+        )
+        import os
+        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}',