Update app.py

app.py

@@ -3,28 +3,28 @@ import gradio as gr
 
 from demo.infer import LiveCCDemoInfer
 
-class GradioBackend:
-    waiting_video_response = 'Waiting for video input...'
-    not_found_video_response = 'Video does not exist...'
-    mode2api = {
-        'Real-Time Commentary': 'live_cc',
-        'Conversation': 'video_qa'
-    }
+with gr.Blocks() as demo:
 
-    @spaces.GPU
-    def __init__(self, model_path: str = 'chenjoya/LiveCC-7B-Instruct'):
-        os.system('pip install flash-attn --no-build-isolation')
-        self.infer = LiveCCDemoInfer(model_path)
-        from kokoro import KPipeline
-        self.audio_pipeline = KPipeline(lang_code='a')
+    class GradioBackend:
+        waiting_video_response = 'Waiting for video input...'
+        not_found_video_response = 'Video does not exist...'
+        mode2api = {
+            'Real-Time Commentary': 'live_cc',
+            'Conversation': 'video_qa'
+        }
 
-    @spaces.GPU
-    def __call__(self, query: str = None, state: dict = {}, mode: str = 'Real-Time Commentary', **kwargs):
-        return getattr(self.infer, self.mode2api[mode])(query=query, state=state, **kwargs)
+        @spaces.GPU
+        def __init__(self, model_path: str = 'chenjoya/LiveCC-7B-Instruct'):
+            self.infer = LiveCCDemoInfer(model_path)
+            from kokoro import KPipeline
+            self.audio_pipeline = KPipeline(lang_code='a')
 
-gradio_backend = GradioBackend() 
+        @spaces.GPU
+        def __call__(self, query: str = None, state: dict = {}, mode: str = 'Real-Time Commentary', **kwargs):
+            return getattr(self.infer, self.mode2api[mode])(query=query, state=state, **kwargs)
+
+    gradio_backend = GradioBackend() 
 
-with gr.Blocks() as demo:
     gr.Markdown("## LiveCC Real-Time Commentary and Conversation - Gradio Demo")
     gr.Markdown("### [LiveCC: Learning Video LLM with Streaming Speech Transcription at Scale (CVPR 2025)](https://showlab.github.io/livecc/)")
     gr.Markdown("1️⃣ Select Mode, Real-Time Commentary (LiveCC) or Conversation (Common QA)")