Use native Gradio inference endpoints

README.md

@@ -4,6 +4,7 @@ emoji: 💊
 colorFrom: blue
 colorTo: indigo
 sdk: gradio
+sdk_version: 6.2.0
 python_version: "3.12"
 app_file: app.py
 pinned: false

app.py

@@ -1,15 +1,12 @@
-"""Public HTTP surface for private-model PFF-PK inference on Hugging Face."""
+"""Public Gradio API for private-model PFF-PK inference."""
 
 from __future__ import annotations
 
-import asyncio
 import os
 from pathlib import Path
 
-from fastapi import FastAPI, HTTPException, Request
-from fastapi.middleware.cors import CORSMiddleware
+import gradio as gr
 from huggingface_hub import hf_hub_download
-from starlette.concurrency import run_in_threadpool
 
 ROOT = Path(__file__).resolve().parent
 MODEL_REPO_ID = os.environ.get("MODEL_REPO_ID", "").strip()
@@ -39,31 +36,9 @@ os.environ.setdefault("PFF_CPU_THREADS", "2")
 
 from services.inference.pff_service import RUNTIME, cached_inference  # noqa: E402
 
-app = FastAPI(
-    title="PFF-PK inference API",
-    description="Server-side zero-shot PK cohort generation and dose intervention.",
-    version="1.0.0",
-)
-app.add_middleware(
-    CORSMiddleware,
-    allow_origins=["https://dfaroughy.github.io"],
-    allow_origin_regex=r"^https?://(localhost|127\.0\.0\.1)(:\d+)?$",
-    allow_methods=["GET", "POST", "OPTIONS"],
-    allow_headers=["content-type"],
-)
-inference_slots = asyncio.Semaphore(1)
-
-
-@app.get("/")
-def root() -> dict[str, str]:
-    return {
-        "service": "PFF-PK inference API",
-        "dashboard": "https://dfaroughy.github.io/pff-pk-dashboard/empirical/",
-    }
 
-
-@app.get("/health")
 def health() -> dict[str, str | bool]:
+    """Return only non-sensitive deployment metadata."""
     metadata = RUNTIME.metadata()
     return {
         "ready": bool(metadata["ready"]),
@@ -73,25 +48,40 @@ def health() -> dict[str, str | bool]:
     }
 
 
-@app.post("/inference")
-async def inference(request: Request) -> dict:
-    content_length = request.headers.get("content-length")
-    if content_length and int(content_length) > 10_000_000:
-        raise HTTPException(status_code=413, detail="inference request is too large")
-    try:
-        payload = await request.json()
-    except Exception as error:
-        raise HTTPException(status_code=400, detail="request body must be JSON") from error
+def inference(payload: dict) -> dict:
+    """Generate a PK cohort from a validated dashboard request."""
+    if not isinstance(payload, dict):
+        raise gr.Error("The inference request must be a JSON object")
     try:
-        async with inference_slots:
-            return await run_in_threadpool(cached_inference, payload)
+        return cached_inference(payload)
     except (ValueError, KeyError, TypeError, FileNotFoundError) as error:
-        raise HTTPException(status_code=400, detail=str(error)) from error
+        raise gr.Error(str(error)) from error
     except Exception as error:
-        raise HTTPException(status_code=500, detail="PFF inference failed") from error
-
+        raise gr.Error("PFF inference failed") from error
+
+
+with gr.Blocks(title="PFF-PK inference API") as demo:
+    gr.Markdown(
+        "# PFF-PK inference API\n"
+        "Server-side zero-shot PK cohort generation for the "
+        "[PFF dashboard](https://dfaroughy.github.io/pff-pk-dashboard/empirical/)."
+    )
+    request = gr.JSON(label="Inference request", value={})
+    result = gr.JSON(label="Inference response")
+    run = gr.Button("Run inference", variant="primary")
+    run.click(
+        inference,
+        inputs=request,
+        outputs=result,
+        api_name="inference",
+        concurrency_limit=1,
+        concurrency_id="pff-inference",
+    )
+    status = gr.JSON(label="Service status")
+    check = gr.Button("Check service")
+    check.click(health, inputs=None, outputs=status, api_name="health")
+
+demo.queue(max_size=32, default_concurrency_limit=1)
 
 if __name__ == "__main__":
-    import uvicorn
-
-    uvicorn.run(app, host="0.0.0.0", port=7860)
+    demo.launch()

requirements.txt

@@ -1,7 +1,6 @@
-fastapi>=0.116,<1
+gradio>=6,<7
 huggingface-hub>=1.0,<2
 lightning>=2.6,<2.7
 numpy>=1.26,<3
 pyyaml>=6,<7
 torch==2.8.0
-uvicorn>=0.35,<1