README.md

---
pipeline_tag: image-classification
license: mit
library_name: pytorch
tags: [pytorch, minimal, demo]
model-index:
  - name: tiny-digits-cnn
    results:
      - task: {type: image-classification}
        metrics:
          - type: accuracy
            value: 0.00   # demo-only (untrained)
---

# Tiny Digits CNN (demo-only)

Toy 28×28 grayscale classifier (10 classes 0–9) for **Hugging Face deployment tests**.
No training—weights are randomly initialized just to validate repo layout, Spaces, and Inference Endpoints.

## Files
- `model.py` — tiny CNN
- `model.safetensors` — weights (create with `python generate_weights.py`)
- `inference.py` — load → preprocess → predict
- `handler.py` — Endpoint handler (`EndpointHandler`)
- `app.py` — Gradio Space UI
- `requirements.txt`, `.gitattributes`, `LICENSE`

## Quickstart (local)
```bash
pip install -r requirements.txt
python generate_weights.py
python app.py

Call via Hosted Inference API (if enabled) or Endpoint
# Replace with your endpoint URL or model API URL
API=https://api-inference.huggingface.co/models/ORG/REPO
curl -X POST "$API" \
  -H "Authorization: Bearer $HF_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"inputs": {"image_base64": "<PUT_BASE64_IMAGE_HERE>"}}'