fix: use demo.launch() for ZeroGPU compliance + attach REST routes to gr_app

app.py

@@ -1,16 +1,12 @@
+import spaces
 import os
-import uvicorn
 import gradio as gr
-import torch
 from transformers import pipeline
-from fastapi import FastAPI, HTTPException
+from fastapi import HTTPException
 from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel, ConfigDict
 from typing import List
 
-import spaces
-
-
 # ── Model ──────────────────────────────────────────────────────────────────────
 ner = pipeline(
     "ner",
@@ -20,7 +16,7 @@ ner = pipeline(
 )
 
 
-# ── Core logic (shared by Gradio UI + REST API) ────────────────────────────────
+# ── Core inference (ZeroGPU) ───────────────────────────────────────────────────
 @spaces.GPU
 def detect_pii(text: str):
     """Run NER and return structured PII results."""
@@ -47,10 +43,40 @@ def detect_pii(text: str):
     return {"text": text, "entities": highlighted}
 
 
-# ── FastAPI app ────────────────────────────────────────────────────────────────
-app = FastAPI(title="PII Detector API", version="1.0.0")
+# ── Gradio UI ──────────────────────────────────────────────────────────────────
+demo = gr.Interface(
+    fn=detect_pii,
+    inputs=gr.Textbox(
+        label="Input Text",
+        placeholder="Enter Sinhala-English codemixed text...",
+        lines=4,
+    ),
+    outputs=gr.HighlightedText(label="Detected PII Entities"),
+    title="Sinhala-English Codemix PII Detector",
+    description=(
+        "Detects PII entities (PERSON, ORG, PHONE, EMAIL, NIC, etc.) in "
+        "Sinhala-English code-mixed text using the Ramosh99/mixcode_pii_detector model.\n\n"
+        "**REST API:** POST `/api/detect` with `{\"text\": \"...\"}` · Docs: `/api/docs`"
+    ),
+    examples=[
+        ["Mage name Kamal, man wada karanne MAS Holdings eke, phone 071-1234567"],
+        ["මගේ email address kamal.perera@gmail.com, NIC number 199912345678"],
+        ["I live in Colombo and my phone number is 0771234567"],
+    ],
+    flagging_mode="never",
+)
 
-app.add_middleware(
+# ── Launch Gradio (non-blocking) — properly initialises ZeroGPU ───────────────
+# demo.launch() returns Gradio's internal FastAPI app which we extend below.
+gr_app, _local_url, _share_url = demo.launch(
+    server_name="0.0.0.0",
+    server_port=7860,
+    prevent_thread_lock=True,
+    show_api=False,
+)
+
+# ── REST API routes added to Gradio's FastAPI app ─────────────────────────────
+gr_app.add_middleware(
     CORSMiddleware,
     allow_origins=["*"],
     allow_methods=["*"],
@@ -60,11 +86,7 @@ app.add_middleware(
 
 class DetectRequest(BaseModel):
     model_config = ConfigDict(
-        json_schema_extra={
-            "example": {
-                "text": "Mage name Kamal, phone 071-1234567"
-            }
-        }
+        json_schema_extra={"example": {"text": "Mage name Kamal, phone 071-1234567"}}
     )
     text: str
 
@@ -82,20 +104,18 @@ class DetectResponse(BaseModel):
     redacted_text: str
 
 
-@app.get("/api/health")
+@gr_app.get("/api/health")
 def health_check():
     """Health check endpoint."""
     return {"status": "ok", "model": "Ramosh99/mixcode_pii_detector"}
 
 
-@app.post("/api/detect", response_model=DetectResponse)
+@gr_app.post("/api/detect", response_model=DetectResponse)
 def api_detect_pii(request: DetectRequest):
     """
-    Detect PII entities in the provided text.
-
-    - **text**: Input Sinhala-English code-mixed text
+    Detect PII entities in Sinhala-English code-mixed text.
 
-    Returns detected entities with their positions, confidence scores,
+    Returns detected entities with positions, confidence scores,
     and a redacted version of the original text.
     """
     if not request.text.strip():
@@ -103,7 +123,6 @@ def api_detect_pii(request: DetectRequest):
 
     result = detect_pii(request.text)
 
-    # Build redacted text by replacing entity spans with [ENTITY_TYPE]
     text = result["text"]
     entities = sorted(result["entities"], key=lambda e: e["start"], reverse=True)
     redacted = text
@@ -117,31 +136,5 @@ def api_detect_pii(request: DetectRequest):
     }
 
 
-# ── Gradio UI ──────────────────────────────────────────────────────────────────
-demo = gr.Interface(
-    fn=detect_pii,
-    inputs=gr.Textbox(
-        label="Input Text",
-        placeholder="Enter Sinhala-English codemixed text...",
-        lines=4,
-    ),
-    outputs=gr.HighlightedText(label="Detected PII Entities"),
-    title="Sinhala-English Codemix PII Detector",
-    description=(
-        "Detects PII entities (PERSON, ORG, PHONE, EMAIL, NIC, etc.) in "
-        "Sinhala-English code-mixed text using the Ramosh99/mixcode_pii_detector model.\n\n"
-        "**REST API:** POST `/api/detect` with `{\"text\": \"...\"}` · Docs: `/docs`"
-    ),
-    examples=[
-        ["Mage name Kamal, man wada karanne MAS Holdings eke, phone 071-1234567"],
-        ["මගේ email address kamal.perera@gmail.com, NIC number 199912345678"],
-        ["I live in Colombo and my phone number is 0771234567"],
-    ],
-    flagging_mode="never",
-)
-
-# Mount Gradio inside FastAPI so both the UI and the API are served together
-app = gr.mount_gradio_app(app, demo, path="/")
-
-# Start the ASGI server — keeps the process alive on HF Spaces
-uvicorn.run(app, host="0.0.0.0", port=7860)
+# ── Keep the process alive ─────────────────────────────────────────────────────
+demo.block_thread()

requirements.txt

@@ -3,5 +3,4 @@ torch>=2.2.0
 sentencepiece>=0.2.0
 huggingface_hub>=0.19.3,<0.26.0
 fastapi>=0.110.0
-uvicorn>=0.29.0
 pydantic>=2.0.0