Add Support for Ovis2.5 Multi-Modal Model​

[xschen-beb]

Model description

Key Features:

Small Model Performance: Optimized training strategies enable small-scale models to achieve higher capability density, demonstrating cross-tier leading advantages.

Enhanced Reasoning Capabilities: Significantly strengthens Chain-of-Thought (CoT) reasoning abilities through the combination of instruction tuning and preference learning.

Video and Multi-Image Processing: Video and multi-image data are incorporated into training to enhance the ability to handle complex visual information across frames and images.

Multilingual Support and OCR: Enhances multilingual OCR beyond English and Chinese and improves structured data extraction from complex visual elements like tables and charts.

Open source status

• The model implementation is available
• The model weights are available

Provide useful links for the implementation

[1] Arxiv: https://arxiv.org/abs/2508.11737
[2] Huggingface: https://huggingface.co/AIDC-AI/Ovis2-8B

[Rocketknight1]

I think we could definitely support a PR for this! cc @zucchini-nlp

[zucchini-nlp]

We have a PR adding it in #40321

[xschen-beb]

We have a PR adding it in #40321

Thank you for your response. However, we were unable to find the Ovis2_5 model under the original transformers/models directory. By the way, regarding the Ovis2 model, as it is based on the Qwen2.5-vl architecture, in addition to using the Ovis2 under transformers/models, would it also be possible to use Qwen2.5-vl? Thank you for your clarification.

[zucchini-nlp]

we were unable to find the Ovis2_5 model under the original transformers/models directory

See the linked PR, it is still under progress and ig it will take some time before we can add the model. The contributor has been inactive for a while...

would it also be possible to use Qwen2.5-vl

The Qwen is a separate model and is supported in transformers. See this documentation for supported models and their usage, it should help to find the suitable model for your use-case

[mahimairaja]

@zucchini-nlp are you guys still working on the PR? Need help?

[zucchini-nlp]

asked the contributor under PR. If they need any help, I will tag you as well @mahimairaja, thanks 🤗