---
title: Wafer Defect Model API
emoji: 🌀
colorFrom: blue
colorTo: indigo
sdk: docker
app_port: 7860
pinned: false
---

# Wafer Defect Model API (remote inference backend)

Hosts `radai-agent/radai-wm811k-defect-detection` (RadAI WM-811K ResNet34)
on Hugging Face's infrastructure and exposes it as a small JSON API, so your
local wafer defect tool doesn't need torch/torchvision installed.

## Endpoints

- `GET /` — health check
- `POST /predict_map` — body: `{"wafer_map": [[...], [...], ...]}` (2D array,
  any size, raw grayscale or bin values). Returns:
  ```json
  {
    "predictions": [{"label": "Edge-Ring", "desc": "...", "prob": 0.87}, ...],
    "preview_raw": [[...64x64, 0-1 normalized...]]
  }
  ```

## Deploying

1. On huggingface.co, create a new Space -> SDK: **Docker**, hardware: free CPU basic.
2. Push these files (`app.py`, `requirements.txt`, `Dockerfile`, this `README.md`)
   to the Space's git repo.
3. Wait for the build to finish, then note the Space URL, e.g.
   `https://YOUR-USERNAME-wafer-defect-model.hf.space`.
4. Free-tier Spaces sleep after inactivity and take ~20-30s to wake on the
   first request afterward — expected, not a bug.