fix

app.py

@@ -1,3 +1,5 @@
+import os
+os.system('pip install flash-attn --no-build-isolation')
 import gradio as gr
 
 from demo.infer import LiveCCDemoInfer
@@ -25,7 +27,7 @@ with gr.Blocks() as demo:
     gr.Markdown("1️⃣ Select Mode, Real-Time Commentary (LiveCC) or Conversation (Common QA)")
     gr.Markdown("2️⃣🅰️ Real-Time Commentary:  Input a query (optional) -> Click or upload a video.")
     gr.Markdown("2️⃣🅱️ Conversation: Click or upload a video -> Input a query.")
-    gr.Markdown("*Web Gradio has unexpected latency (3s~5s). If you want to enjoy the very real-time experience, please try our CLI demo https://github.com/showlab/livecc*")
+    gr.Markdown("*Web Gradio has unexpected latency (3s~5s). If you want to enjoy the very real-time experience, please deploy locally https://github.com/showlab/livecc*")
     gr_state = gr.State({}, render=False) # control all useful state, including kv cache
     gr_video_state = gr.JSON({}, visible=False) # only record video state, belong to gr_state but lightweight
     gr_static_trigger = gr.Number(value=0, visible=False) # control start streaming or stop
@@ -45,10 +47,7 @@ with gr.Blocks() as demo:
             gr_examples = gr.Examples(
                 examples=[
                     'demo/sources/howto_fix_laptop_mute_1080p.mp4',
-                    # 'demo/sources/warriors_vs_rockets_2025wcr1_mute_1080p.mp4',
-                    # 'demo/sources/dota2_facelessvoid_mute_1080p.mp4',
-                    # 'demo/sources/spacex_falcon9_mute_1080p.mp4',
-                    # 'demo/sources/writing_mute_1080p.mp4'
+                    'demo/sources/writing_mute_1080p.mp4'
                 ],
                 inputs=[gr_video],
             )
@@ -59,8 +58,6 @@ with gr.Blocks() as demo:
                 gr_radio_mode = gr.Radio(label="Select Mode", choices=["Real-Time Commentary", "Conversation"], elem_id="gr_radio_mode", value='Real-Time Commentary', interactive=True) 
 
             def gr_chatinterface_fn(message, history, state, video_path, mode):
-                if mode == 'Real-Time Commentary':
-                    return gradio_backend.waiting_video_response, state
                 state['video_path'] = video_path
                 response, state = gradio_backend(query=message, state=state, mode=mode)
                 return response, state