Upload app.py with huggingface_hub

app.py

@@ -388,5 +388,27 @@ def finale(code: str, language: str = "es", variant: float = 0) -> str:
     yield from _stream_fixed(finale_text(lang), char_delay=0.03)
 
 
+def _warm_up():
+    """Load the model and run one tiny generation at startup, in the background,
+    so the first visitor (a judge) doesn't pay the cold-start penalty: the GGUF
+    is already in memory and llama.cpp is warm before anyone turns the dial."""
+    try:
+        canonical = nearest_station_frequency(95.0) or 95.0
+        for _ in stream_broadcast(
+            station_system_prompt(canonical, "es"),
+            user_prompt="hola",
+            seed=station_seed(canonical),
+            max_tokens=4,
+        ):
+            pass
+        print("[warmup] model preloaded and warm")
+    except Exception as exc:  # noqa: BLE001 - never let warm-up crash boot
+        print(f"[warmup] skipped: {exc}")
+
+
+# Kick the warm-up as soon as the module is imported (HF imports app.py on boot).
+threading.Thread(target=_warm_up, daemon=True).start()
+
+
 if __name__ == "__main__":
     app.launch(server_name="0.0.0.0", server_port=7860, show_error=True, ssr_mode=False)