---
title: OvisOCR2
canonical_url: "https://www.modelscope.cn/studios/ATH-MaaS/OvisOCR2"
md_url: "https://www.modelscope.cn/studios/ATH-MaaS/OvisOCR2.md"
repository: ATH-MaaS/OvisOCR2
chinese_name: "OvisOCR2 文档解析"
last_updated: 2026-07-22
sdk_type: gradio
sdk_version: 6.17.3
downloads: 0
stars: 68
---

# OvisOCR2

> OvisOCR2 - ATH-MaaS 在 ModelScope 创建的在线 Demo。OvisOCR2 image and PDF document parsing demo with streaming Markdown output.

ATH-MaaS/OvisOCR2 是 ModelScope 魔搭社区上的在线可交互 Demo（创空间），基于 gradio 6.17.3 构建，中文名为「OvisOCR2 文档解析」。

- **Repository**: ATH-MaaS/OvisOCR2
- **SDK**: gradio
- **SDK version**: 6.17.3
- **Downloads**: 0
- **Stars**: 68
- **Last updated**: 2026-07-22

Source: https://www.modelscope.cn/studios/ATH-MaaS/OvisOCR2

---

# OvisOCR2 — Document OCR Demo

OvisOCR2 is a compact 0.8B end-to-end model for page-level document parsing.
This Studio accepts images and PDFs and streams structured Markdown containing
text, LaTeX formulas, HTML tables, and cropped visual regions.

PDF pages are rasterized and inferred independently in Base mode. Up to four
pages share one xGPU request by default, reducing repeated scheduling overhead
while preserving per-page results and navigation in the browser.

Model: [ATH-MaaS/OvisOCR2](https://modelscope.cn/models/ATH-MaaS/OvisOCR2)

## Runtime

- Gradio 6.17.3
- Ada xGPU with 48 GB VRAM
- ModelScope model cache under `/mnt/workspace/modelscope-cache`
- `app.py` serves the pre-built frontend in `dist/`; Node.js is not required at runtime

## Configuration

- `OVISOCR_MODEL_PROVIDER=modelscope`
- `MODELSCOPE_CACHE=/mnt/workspace/modelscope-cache`
- `OVISOCR_MAX_NEW_TOKENS=16384`
- `OVISOCR_MAX_PDF_PAGES=50`
- `OVISOCR_PAGES_PER_GPU_REQUEST=4`