# ATH-MaaS/Ovis

A novel Multimodal Large Language Model (MLLM) architecture, designed to structurally align visual and textual embeddings.

Repository: https://github.com/ATH-MaaS/Ovis
Canonical: https://ross.abutalabs.com/products/ovis
Homepage: https://huggingface.co/AIDC-AI/Ovis2.5-9B
Language: Python
License: Apache-2.0
License Family: permissive
Topics: chatbot, llama3, multimodal, multimodal-large-language-models, multimodality, qwen, vision-language-learning, vision-language-model
Last push: 2026-07-15T14:04:16+00:00

## Health v2 (maintenance only)
Score: 65/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 92, release rhythm 35, longevity 57
- inputs: {"age_days": 811, "days_push": 49, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1512, forks 89 (observed 2026-08-28T04:04:56.127101+00:00)

## What it is
Ovis is an open-source Multimodal Large Language Model (MLLM) architecture that structurally aligns visual and textual embeddings, with releases like Ovis2.5 featuring native-resolution vision (NaViT) and an optional reflective 'thinking mode'. It ships model weights (2B-34B, quantized variants) and Python code for inference and fine-tuning on Hugging Face Transformers.

## Use cases
- run a vision-language model for image question answering
- extract and understand charts, tables, and documents with OCR
- analyze videos and multi-image inputs with an MLLM
- deploy a small multimodal LLM on resource-constrained hardware
- fine-tune an open-source MLLM for custom multimodal tasks
- add visual grounding and image reasoning to an application

## When to choose
- you need an open-source MLLM with strong chart/document/OCR performance at small scale
- you want native-resolution image understanding without lossy tiling
- you need an optional deep-reasoning 'thinking mode' trading latency for accuracy
- you want Apache-2.0 licensed weights with quantized variants

## When to avoid
- you only need text-only LLM inference
- you need a fully managed multimodal API rather than self-hosting models
- you lack GPU resources for even the 2B model
- you need audio or speech multimodality, which Ovis does not cover

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-inference, ocr, image-processing, chatbot
- domain: large-language-models, computer-vision, artificial-intelligence, deep-learning
- platform: python, cross-platform
- tags: multimodal, vision-language-model, mllm, native-resolution, thinking-mode, video-understanding, apache-2.0, natural-language-processing, gpu, linux

## Member repositories
- ATH-MaaS/Ovis (main) score 65

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:56.127101+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:32:17.922560+00:00, confidence not recorded.
  - readme: https://github.com/ATH-MaaS/Ovis (fetched 2026-08-28T04:04:56.127101+00:00, sha 67668bef415e)
  - homepage: https://huggingface.co/AIDC-AI/Ovis2.5-9B (fetched 2026-08-29T11:36:12.587408+00:00, sha 77215d91509c)
  - site_page: https://huggingface.co/docs (fetched 2026-08-29T11:36:12.597141+00:00, sha bdec26667b98)
  - site_page: https://huggingface.co/docs/safetensors (fetched 2026-08-29T11:36:12.601231+00:00, sha 7fa2c18b9c26)
  - site_page: https://huggingface.co/docs/inference-providers (fetched 2026-08-29T11:36:12.602949+00:00, sha 8a5d0f819473)
  - site_page: https://huggingface.co/docs/hub/model-cards (fetched 2026-08-29T11:36:12.605039+00:00, sha 60ded09a56b0)
  - site_page: https://huggingface.co/pricing (fetched 2026-08-29T11:36:12.599097+00:00, sha de6b7a178be5)
  - site_page: https://huggingface.co/huggingface (fetched 2026-08-29T11:36:12.607334+00:00, sha e8292d1ffb39)
- Data as of 2026-08-30T08:39:29.467469+00:00.