# EvolvingLMMs-Lab/LLaVA-OneVision-2

Fully Open Framework for Democratized Multimodal Training

Repository: https://github.com/EvolvingLMMs-Lab/LLaVA-OneVision-2
Canonical: https://ross.abutalabs.com/products/llava-onevision-2
Homepage: https://evolvinglmms-lab.github.io/LLaVA-OneVision-2/projects/index.html
Language: Python
License: Apache-2.0
License Family: permissive
Topics: llava, llm, mllm, qwen3, vision-language-model, llava-onevision
Last push: 2026-09-02T03:16:51+00:00

## Health v2 (maintenance only)
Score: 72/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 100, release rhythm 64, longevity 25
- inputs: {"age_days": 351, "days_push": 0, "days_rel": 27, "gap_med": 223, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1195, forks 80 (observed 2026-09-03T02:15:05.796555+00:00)

## What it is
A fully open framework for training multimodal large language models, releasing models, datasets, and training recipes for the LLaVA-OneVision family. It includes an 8B-class video MLLM with codec-aligned dense video input and a codec-aligned vision encoder.

## Use cases
- train a vision-language model from scratch
- build a video understanding MLLM
- reproduce open multimodal model training
- fine-tune LLaVA-OneVision on my own data
- get an open alternative to Qwen-VL
- train a vision encoder for image and video
- run reinforcement learning post-training for multimodal reasoning

## When to choose
- you need fully reproducible open weights, data, and recipes for multimodal training
- you want strong video comprehension on long footage with an 8B model
- you want an efficient codec-aligned vision encoder that uses fewer visual tokens

## When to avoid
- you only need to run inference with a hosted API
- you lack GPU resources for large-scale multimodal training
- you need a small lightweight model for edge devices

## Facets
- artifact type: framework
- maturity: active
- function: machine-learning, deep-learning, llm-training, image-processing, video-processing, nlp
- domain: large-language-models, machine-learning, computer-vision, deep-learning, artificial-intelligence
- platform: python
- tags: vision-language-model, multimodal, llava, qwen3, video-understanding, open-models, model-training, gpu, linux, docker

## Member repositories
- EvolvingLMMs-Lab/LLaVA-OneVision-2 (main) score 72

## Provenance
- Observed fields: from GitHub, fetched 2026-09-03T02:15:05.796555+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 2026-08-30T06:22:27.056240+00:00, confidence not recorded.
  - readme: https://github.com/EvolvingLMMs-Lab/LLaVA-OneVision-2 (fetched 2026-09-03T02:15:05.796555+00:00, sha c7dc7a3cd3d5)
  - homepage: https://evolvinglmms-lab.github.io/LLaVA-OneVision-2/projects/index.html (fetched 2026-08-29T12:29:50.529083+00:00, sha c715afaacfd7)
- Data as of 2026-08-30T08:39:29.467469+00:00.