fix: export single ASGI app for Hugging Face supervisor on ZeroGPU

README.md

@@ -6,7 +6,6 @@ colorTo: blue
 sdk: gradio
 sdk_version: 5.16.0
 app_file: app.py
-app_port: 7860
 pinned: false
 ---
 

app.py

@@ -1,25 +1,41 @@
-import spaces  # ZeroGPU MUST be imported on line 1
-import os
-import uvicorn
+try:
+    import spaces
+    has_spaces = True
+except ImportError:
+    has_spaces = False
+
 import gradio as gr
 from backend.app.main import app as fastapi_app
 from backend.app.ml.manager import model_manager
 
 # 1. ZeroGPU Inference Function
-@spaces.GPU
-def predict_ner(text: str):
-    if not text or not text.strip():
-        return "Please provide clinical text."
-    try:
-        model = model_manager.get_ner_model()
-        results = model.predict(text)
-        if not results:
-            return "No biomedical entities detected."
-        return "\n".join([f"• [{e.label}] {e.text} (Confidence: {round((e.confidence or 1.0)*100, 1)}%)" for e in results])
-    except Exception as e:
-        return f"Inference notice: {e}"
+if has_spaces:
+    @spaces.GPU
+    def predict_ner(text: str):
+        if not text or not text.strip():
+            return "Please provide clinical text."
+        try:
+            model = model_manager.get_ner_model()
+            results = model.predict(text)
+            if not results:
+                return "No biomedical entities detected."
+            return "\n".join([f"• [{e.label}] {e.text} (Confidence: {round((e.confidence or 1.0)*100, 1)}%)" for e in results])
+        except Exception as e:
+            return f"Inference notice: {e}"
+else:
+    def predict_ner(text: str):
+        if not text or not text.strip():
+            return "Please provide clinical text."
+        try:
+            model = model_manager.get_ner_model()
+            results = model.predict(text)
+            if not results:
+                return "No biomedical entities detected."
+            return "\n".join([f"• [{e.label}] {e.text} (Confidence: {round((e.confidence or 1.0)*100, 1)}%)" for e in results])
+        except Exception as e:
+            return f"Inference notice: {e}"
 
-# 2. Gradio Interface (Provides the @spaces.GPU event hook for ZeroGPU supervisor)
+# 2. Gradio Interface (Provides the @spaces.GPU hook for ZeroGPU supervisor)
 with gr.Blocks(title="SanjeevaniAI Healthcare Intelligence") as demo:
     gr.Markdown("# 🏥 SanjeevaniAI — Healthcare Intelligence Engine")
     gr.Markdown(
@@ -34,9 +50,6 @@ with gr.Blocks(title="SanjeevaniAI Healthcare Intelligence") as demo:
     btn = gr.Button("⚡ Run ZeroGPU Clinical NER", variant="primary")
     btn.click(fn=predict_ner, inputs=inp, outputs=out)
 
-# 3. Mount Gradio under /ui so FastAPI controls root / and /api/v1/ with ZERO SvelteKit interference
-app = gr.mount_gradio_app(fastapi_app, demo, path="/ui")
-
-if __name__ == "__main__":
-    port = int(os.environ.get("PORT", 7860))
-    uvicorn.run(app, host="0.0.0.0", port=port)
+# 3. Mount Gradio under /gradio so FastAPI controls root / and /api/v1/ with ZERO SvelteKit interference
+# Hugging Face's supervisor (Process [1]) automatically serves this exported `app` on port 7860!
+app = gr.mount_gradio_app(fastapi_app, demo, path="/gradio")