Run inference on CPU with ZeroGPU probe

app.py

@@ -129,7 +129,11 @@ def health(api_key: str = "") -> str:
     )
 
 
-@spaces.GPU(duration=300)
+@spaces.GPU(duration=10)
+def zerogpu_probe() -> str:
+    return "ok"
+
+
 def transcribe_b64(
     api_key: str,
     audio_base64: str,
@@ -173,7 +177,6 @@ def transcribe_b64(
     )
 
 
-@spaces.GPU(duration=300)
 def speech_b64(api_key: str, text: str, voice: str = "zf_xiaoxiao", speed: float = 1.0) -> str:
     check_key(api_key)
     speed = float(speed)
@@ -212,6 +215,12 @@ with gr.Blocks(title="Voice API") as demo:
     with gr.Tab("Health"):
         health_out = gr.Textbox(label="result")
         gr.Button("Check").click(health, inputs=[api_key], outputs=[health_out], api_name="health")
+        gr.Button("ZeroGPU probe", visible=False).click(
+            zerogpu_probe,
+            inputs=[],
+            outputs=[health_out],
+            api_name="zerogpu_probe",
+        )
     with gr.Tab("ASR"):
         audio_b64 = gr.Textbox(label="audio_base64", lines=5)
         filename = gr.Textbox(label="filename", value="audio.wav")