Run public inference on CPU Basic

README.md

@@ -17,6 +17,9 @@ Server-side inference for the public PFF-PK pharmacokinetic dashboard. Model
 weights are loaded from a private, immutable Hugging Face model release and are
 never included in this Space repository or returned by its API.
 
+The service runs on CPU Basic hardware. It does not use ZeroGPU and therefore
+does not consume visitors' daily GPU quota.
+
 The named Gradio endpoints are `/health` and `/inference`. Inference is
 serialized to keep memory use bounded; identical requests reuse a response
 keyed by the request and immutable model/configuration fingerprints.

app.py

@@ -6,7 +6,6 @@ import os
 from pathlib import Path
 
 import gradio as gr
-import spaces
 from huggingface_hub import hf_hub_download
 
 ROOT = Path(__file__).resolve().parent
@@ -49,7 +48,6 @@ def health() -> dict[str, str | bool]:
     }
 
 
-@spaces.GPU(duration=120)
 def inference(payload: dict) -> dict:
     """Generate a PK cohort from a validated dashboard request."""
     if not isinstance(payload, dict):