app.py

import json
import os
import sys
from pathlib import Path

import spaces


ROOT = Path(__file__).resolve().parent
RUNTIME_ROOT = Path(os.environ.get("CAPSTONE_RUNTIME_ROOT", "/tmp/call-qa"))
DATA_ROOT = RUNTIME_ROOT / "data"

os.environ.setdefault("CAPSTONE_DATA_ROOT", str(DATA_ROOT))
os.environ.setdefault("CAPSTONE_EVAL_RESULTS", str(RUNTIME_ROOT / "evaluation_results"))
os.environ.setdefault("CAPSTONE_FRONTEND_PUBLIC", str(RUNTIME_ROOT / "frontend_public"))
os.environ.setdefault("CAPSTONE_ENV_FILE", str(ROOT / ".env"))
os.environ.setdefault("ENABLE_ACOUSTIC", "0")
os.environ.setdefault("START_WORKER", "1")


def _prepare_runtime() -> None:
    """Create the writable files expected by the existing pipeline modules."""
    manifest = DATA_ROOT / "na_testset" / "manifest.json"
    manifest.parent.mkdir(parents=True, exist_ok=True)
    if not manifest.exists():
        manifest.write_text(json.dumps([]), encoding="utf-8")

    for path in (
        RUNTIME_ROOT / "evaluation_results",
        RUNTIME_ROOT / "frontend_public",
    ):
        path.mkdir(parents=True, exist_ok=True)


_prepare_runtime()
sys.path.insert(0, str(ROOT / "backend"))


@spaces.GPU(duration=10)
def gpu_snapshot():
    """Allocate a tiny tensor to verify that ZeroGPU scheduling is available."""
    import torch

    probe = torch.ones(1, device="cuda")
    return {
        "available": bool(torch.cuda.is_available()),
        "device": torch.cuda.get_device_name(0),
        "probe": float(probe.item()),
    }


from app.space import build_space_app, lifespan  # noqa: E402

demo, server_app = build_space_app(gpu_snapshot)


if __name__ == "__main__":
    port = int(os.environ.get(
        "CAPSTONE_PORT",
        os.environ.get("GRADIO_SERVER_PORT", os.environ.get("PORT", "7860")),
    ))
    demo.launch(
        server_name="0.0.0.0",
        server_port=port,
        ssr_mode=False,
        _app=server_app,
        app_kwargs={"lifespan": lifespan},
    )