---
title: Document Classification Demo
emoji: 🖼️
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 6.20.0
app_file: app.py
pinned: false
license: other
short_description: Classify any document image against labels you choose
---

# Nutrient document classification demo

Open-vocabulary, zero-shot: add candidate classes (a **label** + optional **description** per row), upload a
document page, and the model ranks them — no fixed class list, no per-class training. Two models, one tab each:

- **v2 flagship** — the commercial model (best accuracy).
- **v1 (open-weight)** — a downloadable open-vocab classifier.

→ [model](https://huggingface.co/nutrientdocs/document-classification-v2) ·
[leaderboard](https://huggingface.co/spaces/nutrientdocs/document-classification-leaderboard) ·
[benchmark](https://huggingface.co/datasets/nutrientdocs/document-classification-benchmark)

## Secret required

The **v2** model is commercial and loaded from a **private** repo, so this Space needs an `HF_TOKEN`
**secret** (Settings → Variables and secrets) with read access to it. The weights are used server-side only
and are never downloadable. Without the secret, the demo marks the v2 tab _unavailable_ and the open **v1**
model still runs. Inference runs on **ZeroGPU**.

Sample document images for quick testing are in `examples/`.