Add Gradio interface for ZeroGPU detection - fixes 'No @spaces.GPU function detected' warning

app.py

@@ -378,10 +378,37 @@ def interactive_ui() -> str:
     """
 
 
-app = fastapi_app
+# Gradio interface for ZeroGPU detection - ZeroGPU requires Gradio SDK
+import gradio as gr
 
+@spaces.GPU(duration=300)
+def gradio_generate(
+    prompt: str,
+    max_new_tokens: int = MAX_NEW_TOKENS,
+    temperature: float = DEFAULT_TEMPERATURE,
+    top_p: float = DEFAULT_TOP_P,
+) -> str:
+    """Gradio interface function with GPU decorator for ZeroGPU detection."""
+    return _generate(prompt, max_new_tokens, temperature, top_p)
 
-if __name__ == "__main__":  # pragma: no cover
-    import uvicorn
+# Create Gradio interface - this ensures ZeroGPU detects the GPU function
+gradio_interface = gr.Interface(
+    fn=gradio_generate,
+    inputs=[
+        gr.Textbox(label="Prompt", lines=5, placeholder="Enter your router prompt here..."),
+        gr.Slider(minimum=64, maximum=2048, value=MAX_NEW_TOKENS, step=16, label="Max New Tokens"),
+        gr.Slider(minimum=0.0, maximum=2.0, value=DEFAULT_TEMPERATURE, step=0.05, label="Temperature"),
+        gr.Slider(minimum=0.0, maximum=1.0, value=DEFAULT_TOP_P, step=0.05, label="Top-p"),
+    ],
+    outputs=gr.Textbox(label="Generated Response", lines=10),
+    title="Router Model API - ZeroGPU",
+    description=f"Model: {MODEL_ID} | Strategy: {ACTIVE_STRATEGY or 'pending'}",
+)
 
-    uvicorn.run(app, host="0.0.0.0", port=int(os.environ.get("PORT", 7860)))
+# Mount FastAPI app within Gradio for API endpoints
+# The main app must be Gradio for ZeroGPU detection, with FastAPI mounted for API routes
+app = gr.mount_gradio_app(fastapi_app, gradio_interface, path="/")
+
+
+if __name__ == "__main__":  # pragma: no cover
+    app.launch(server_name="0.0.0.0", server_port=int(os.environ.get("PORT", 7860)))

requirements.txt

@@ -6,3 +6,4 @@ torch>=2.1.0
 transformers>=4.40.0
 uvicorn>=0.22.0
 sentencepiece>=0.1.99
+gradio>=4.0.0