update

app.py

@@ -36,7 +36,7 @@ pipe_dict = {
     k: pipeline(
         "automatic-speech-recognition",
         model=v,
-        device="cuda" if torch.cuda.is_available() or is_hf else "cpu",
+        device="cuda" if torch.cuda.is_available() else "cpu",
     )
     for k, v in model_dict.items()
 }
@@ -82,9 +82,10 @@ def transcribe_galgame_whisper(audio) -> tuple[str, float]:
     return transcribe_common(audio, "galgame-whisper-wip")
 
 
-logger.info("Warm-up...")
-transcribe_large_v3_turbo("test.wav")
-logger.success("Warm-up done!")
+def warmup():
+    logger.info("Warm-up...")
+    return transcribe_large_v3_turbo("test.wav")
+
 
 initial_md = """
 # Galgame-Whisper (WIP) Demo
@@ -143,8 +144,7 @@ with gr.Blocks() as app:
             time_kotoba_v2 = gr.Textbox(label="Time taken")
             output_kotoba_v2 = gr.Textbox(label="Result")
 
-    with gr.Row():
-        refresh_button = gr.Button("Refresh Status")  # Create a refresh button
+    warmup_result = gr.Textbox(label="Warm-up result", visible=False)
 
     button_v2.click(transcribe_large_v2, inputs=audio, outputs=[output_v2, time_v2])
     button_v3.click(transcribe_large_v3, inputs=audio, outputs=[output_v3, time_v3])
@@ -164,4 +164,6 @@ with gr.Blocks() as app:
         inputs=audio,
         outputs=[output_galgame, time_galgame],
     )
+
+    app.load(warmup, inputs=[], outputs=[warmup_result], queue=True)
 app.launch(inbrowser=True)