fix: expose ZeroGPU handler from Space entrypoint

app.py

@@ -3,6 +3,8 @@ 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"))
@@ -33,11 +35,28 @@ def _prepare_runtime() -> None:
 _prepare_runtime()
 sys.path.insert(0, str(ROOT / "backend"))
 
-from app.space import app  # noqa: E402
+
+@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  # noqa: E402
+
+app = build_space_app(gpu_snapshot)
 
 
 if __name__ == "__main__":
     import uvicorn
 
-    port = int(os.environ.get("GRADIO_SERVER_PORT", os.environ.get("PORT", "7860")))
+    default_port = "7861" if os.environ.get("SPACES_ZERO_GPU") else "7860"
+    port = int(os.environ.get("GRADIO_SERVER_PORT", os.environ.get("PORT", default_port)))
     uvicorn.run(app, host="0.0.0.0", port=port)

backend/app/space.py

@@ -1,7 +1,6 @@
 import os
 
 import gradio as gr
-import spaces
 from fastapi.responses import RedirectResponse
 from sqlalchemy import func
 
@@ -32,41 +31,26 @@ def runtime_snapshot():
     return snapshot
 
 
-@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()),
-    }
-
-
-with gr.Blocks(title="Call QA Processing") as demo:
-    gr.Markdown("# Call QA Processing")
-    state = gr.JSON(label="Runtime status")
-    refresh = gr.Button("Refresh", variant="primary")
-    refresh.click(runtime_snapshot, outputs=state, api_name="runtime_status")
-    demo.load(runtime_snapshot, outputs=state)
-    gpu_state = gr.JSON(label="ZeroGPU status")
-    gpu_check = gr.Button("Check ZeroGPU")
-    gpu_check.click(gpu_snapshot, outputs=gpu_state, api_name="gpu_status")
-
-
-# Replace the JSON root with the Space UI while preserving every API route.
-api.router.routes = [
-    route
-    for route in api.router.routes
-    if not (getattr(route, "path", None) == "/" and "GET" in getattr(route, "methods", set()))
-]
-
-
-@api.get("/", include_in_schema=False)
-def space_root():
-    return RedirectResponse(url="/gradio/")
-
-
-app = gr.mount_gradio_app(api, demo, path="/gradio")
+def build_space_app(gpu_snapshot):
+    with gr.Blocks(title="Call QA Processing") as demo:
+        gr.Markdown("# Call QA Processing")
+        state = gr.JSON(label="Runtime status")
+        refresh = gr.Button("Refresh", variant="primary")
+        refresh.click(runtime_snapshot, outputs=state, api_name="runtime_status")
+        demo.load(runtime_snapshot, outputs=state)
+        gpu_state = gr.JSON(label="ZeroGPU status")
+        gpu_check = gr.Button("Check ZeroGPU")
+        gpu_check.click(gpu_snapshot, outputs=gpu_state, api_name="gpu_status")
+
+    # Replace the JSON root with the Space UI while preserving every API route.
+    api.router.routes = [
+        route
+        for route in api.router.routes
+        if not (getattr(route, "path", None) == "/" and "GET" in getattr(route, "methods", set()))
+    ]
+
+    @api.get("/", include_in_schema=False)
+    def space_root():
+        return RedirectResponse(url="/gradio/")
+
+    return gr.mount_gradio_app(api, demo, path="/gradio")