import os # Disable Gradio SSR mode to avoid Node.js sub-process port collisions os.environ["GRADIO_SSR_MODE"] = "False" os.environ.pop("GRADIO_NODE_SERVER_PORT", None) os.environ.setdefault("ALLOW_LOCAL_ORIGINS_IN_PROD", "true") import spaces import gradio as gr import uvicorn import threading import time import urllib.request from app.main import app def _keep_alive_loop(): """Background keep-alive thread pinging public endpoint every 10 minutes to prevent idle sleep.""" time.sleep(30) url = "https://subh-007-allcited-api.hf.space/v1/readyz" while True: try: req = urllib.request.Request(url, headers={"User-Agent": "Allcited-KeepAlive-Internal/1.0"}) with urllib.request.urlopen(req, timeout=15) as resp: print(f"[keep-alive] Self-ping status {resp.status}") except Exception as e: print(f"[keep-alive] Self-ping note: {e}") time.sleep(600) threading.Thread(target=_keep_alive_loop, daemon=True, name="hf-space-keep-alive").start() @spaces.GPU(duration=30) def gpu_inference_function(text: str) -> str: """ZeroGPU event binding hook for Hugging Face Space analyzer.""" return f"Allcited ZeroGPU ready: {text}" # Send ZeroGPU startup report on FastAPI startup @app.on_event("startup") def on_startup_zerogpu(): try: from spaces.zero import client as zero_client from spaces.config import Config as ZeroConfig if getattr(ZeroConfig, "zero_gpu", False): zero_client.startup_report() print("🚀 ZeroGPU startup report transmitted via FastAPI startup hook!") except Exception as e: print(f"ZeroGPU startup hook: {e}") # Throwaway Gradio launch on unused port to satisfy spaces.one_launch hook try: dummy_block = gr.Interface(fn=gpu_inference_function, inputs="text", outputs="text") dummy_block.launch(prevent_thread_lock=True, server_port=7865, quiet=True) dummy_block.close() print("🚀 ZeroGPU one_launch hook triggered successfully!") except Exception as e: print(f"ZeroGPU throwaway launch note: {e}") with gr.Blocks(title="Allcited API") as demo: gr.Markdown("# Allcited API Server\n\nFastAPI backend with grounded citations.") txt_input = gr.Textbox(label="Status Input", value="ready") txt_output = gr.Textbox(label="Status Output") btn = gr.Button("Ping GPU") btn.click(fn=gpu_inference_function, inputs=txt_input, outputs=txt_output) # Mount Gradio interface at root so HF Space iframe renders, while FastAPI routes take priority app = gr.mount_gradio_app(app, demo, path="/") if __name__ == "__main__": port = int(os.environ.get("PORT", "7860")) if os.environ.get("GRADIO_SSR_MODE", "").lower() in ("false", "0") and port == 7861: port = 7860 print(f"Starting server on port {port} (SSR={os.environ.get('GRADIO_SSR_MODE')})") uvicorn.run(app, host="0.0.0.0", port=port)