---
title: CatBCS
sdk: gradio
app_file: app.py
license: mit
---

# CatBCS / FelineBCS Hugging Face Space

This directory contains the minimal files needed to run the FelineBCS Gradio
inference app as a Hugging Face Space. The public website should call this Space
API; Vercel should not run the model.

## Create the Space

1. In Hugging Face, create a new Space named `catbcs`.
2. Set the Space SDK to **Gradio**.
3. Upload or copy these files into the root of the Space repository:
   - `app.py`
   - `felinebcs_predict.py`
   - `requirements.txt`
   - `README.md`
4. Let Hugging Face install dependencies and launch `app.py`.

## Model artifacts

The trained heads are not committed to the GitHub source repository. Provide the
model artifacts to the Space separately.

Default expected layout:

```text
models/
  clip_vitb32_regressor.joblib
  clip_vitb32_clf9.joblib
  clip_vitb32_clf4.joblib
```

If the files live somewhere else in the Space runtime, set:

```text
FELINEBCS_MODEL_DIR=/path/to/models
```

Do not hardcode local machine paths. If the artifacts are missing, the Space
will still start and the app will show a clear "model artifacts unavailable"
message when a prediction is requested.

## Responsible use

CatBCS / FelineBCS is a non-diagnostic research and triage aid. It was trained
on weak, model-generated labels, not veterinary ground truth. It is not
veterinary advice, diagnosis, or a substitute for a licensed veterinarian's
hands-on examination.