Skip to content
Computer vision

OvisOCR3

компактная OCR-мультимодель для документов

2 min read

OvisOCR3 — универсальная мультимодальная модель на 0,8 млрд параметров от ATH-MaaS, построенная на Qwen3.5-0.8B. Она объединяет разбор документов, общий OCR, извлечение ключевой информации, подписи к рисункам и восстановление таблиц между страницами; нужная задача выбирается промптом. По карточке, среди моделей сопоставимого размера OvisOCR3 лидирует во всех пяти доменах.

Ключевые особенности

  • Разбор документов: Markdown с LaTeX и HTML-таблицами, структурированный JSON с bbox/category/content или только layout.
  • Общий OCR: текст, bounding boxes, LaTeX, данные графиков, перевод, классификация и VQA через разные промпты.
  • Извлечение ключевой информации: key-value записи или JSON по заданному шаблону.
  • Подписи к рисункам: локализованные описания и Mermaid для блок-схем.
  • Сквозные таблицы: собирает HTML-таблицы из последовательных страниц с указанием источника.

Пример

Запуск через vLLM:

from PIL import Image
from vllm import LLM, SamplingParams

model = LLM(model=".", tensor_parallel_size=1,
            gpu_memory_utilization=0.8,
            gdn_prefill_backend="triton",
            limit_mm_per_prompt={"image": 2})
page = Image.open("test1.jpg").convert("RGB")
Подробнее Бенчмарки: OmniDocBench1.6 — 96,74; OCRBench — 88,8; OCRBench KIE — 90,50; CrossTabBench Relation F1 — 97,07.
Для инференса нужна GPU-среда с vLLM; в примере — vllm==0.22.1, одна GPU и gpu_memory_utilization=0.8.
Обучение в четыре этапа: continued pretraining, mid-training, on-policy distillation и reinforcement learning.
`infer` возвращает сырой текст без очистки, починки JSON и удаления повторов — постобработку нужно делать отдельно.
Примеры
Discuss Telegram 110 5.5K

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.6 GB · ready files2.2 GB
  • FP8 / INT8recommendedweights 0.8 GB1.4 GB
  • 4-bitweights 0.4 GB1.0 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # ATH-MaaS/OvisOCR3
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download ATH-MaaS/OvisOCR3 --local-dir ~/models/OvisOCR3
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to OvisOCR3
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. Qwen3.5-0.8B-Basefamily root
  2. Qwen3.5-0.8B
  3. OvisOCR3

New posts about versions of Qwen3.5-0.8B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 9All on Hugging Face
Merges 1All on Hugging Face

About the model

ATH-MaaS/OvisOCR3
Author
ATH-MaaS
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

5.5K110

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments