evolvinglmms-lab/llava-onevision-2

Fully Open Framework for Democratized Multimodal Training

👉 <a href=“asset/xiaohongshu_group.jpeg”>Xiaohongshu Group 👈

🤗 2 Models · Datasets · <a href=“https://arxiv.org/pdf/2605.25979”>Technical Report · <a href=“https://evolvinglmms-lab.github.io/LLaVA-OneVision-2/”>HomePage · <a href=“https://huggingface.co/spaces/FeilongTang/OneVision-Encoder-Codec-View”>Codec Playground · <a href=“https://docs.vllm.ai/en/latest/api/vllm/model_executor/models/llava_onevision2/”>vLLM · <a href=“https://discord.gg/PmdGHMFNP”>Discord

🤗 1.5 Models · Datasets · Technical Report · NeMo · Discord

---

NEWS

• 2026-04-30: Released LLaVA-OneVision-2 — next-generation multimodal model, with new LLaVA-OneVision-2-VideoCaption and LLaVA-OneVision-2-Spatial datasets.
• 2026-02-10: Released OneVision-Encoder — codec-aligned vision encoders, with Technical Report.
• 2025-12-11: Released RL recipe for LLaVA-OneVision-1.5, with Project, Code, Data, and Model.
• 2025-09-30: Released the LLaVA-OneVision-1.5 Technical Report.

Contents

• Introduction
• Results
• Evaluation Reproduction
• Method
• Models
• Datasets
• Quick Start (4B, single node)
• Citation
• Acknowledgement
• LLaVA-OneVision-1.5

Introduction

LLaVA-OneVision-2 is the next-generation release of the LLaVA-OneVision family — a fully open 8B multimodal model that unifies image, long-form video, and spatial understanding under a single architecture, with the entire pipeline (data, encoders, training, checkpoints, logs) released end-to-end.

🎬 Codec-Aligned Vision Encoders

Beyond uniform patchification. OneVision-Encoder and OneVision-Encoder-Lang are HEVC-style vision transformers that add a codec-stream input mode alongside image and uniform-frame video — selecting only motion- and residual-rich patches and sampling dense frames sparsely instead of sparse frames densely. The result is dramatically longer temporal coverage under the same token budget, where prior ViT backbones simply run out of context.

🧊 One Model, Every Modality

Most open multimodal models still live in a 2D, single-image world. LLaVA-OneVision-2-8B-Instruct breaks out of it — one model, native resolution, no task-specific adapters, no hidden tricks.

• Long video — multi-frame reasoning with efficient codec-aligned inference
• 3D-aware spatial reasoning — depth, layout, object relations
• Documents, OCR, charts — structured visual inputs at native resolution

New open-source SOTA across a broad suite of multimodal benchmarks.

🚀 Fully Open, Reproducible from Day One

Four datasets ship with the LLaVA-OneVision family — two new for 2, two carried forward from 1.5:

• LLaVA-OneVision-2-VideoCaption — extremely dense video captions
• LLaVA-OneVision-2-Spatial — 3D-aware spatial reasoning
• LLaVA-OneVision-1.5-Mid-Training-85M — 85M concept-balanced mid-training corpus
• LLaVA-OneVision-1.5-Instruct — full instruction-tuning mixture

And unlike most “open” releases, everything ships alongside them: encoder weights, training code, configs, and full training logs. Reproducible end to end.

Evaluation Results

Evaluation Reproduction

The evaluation settings behind the reported LLaVA-OneVision-2 results are published in the llava-onevision2 branch of lmms-eval:

• Branch: https://github.com/EvolvingLMMs-Lab/lmms-eval/tree/llava-onevision2
• Reproduction guide: https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/llava-onevision2/README.md

That branch includes the exact evaluation model wrapper, benchmark/task configs, Docker environment, and thin launcher scripts used to reproduce the reported numbers for both the frames backend and the codec backend.

Method

Codec-Style Patch Selection

Standard video pipelines uniformly sample a handful of frames and process every patch — most of it static background. We borrow from HEVC: keep I-frames dense, keep only motion- and residual-rich patches from P-frames. Same 54-token budget, 18 frames instead of 6 — 3× the temporal range, no extra LLM context, no input-type adapters.

One Encoder, Every Modality

Most multimodal stacks ship a different tokenizer per input type — one path for images, another for video, a third for multi-image. We don’t. Image, uniform frames, and codec-aligned tokens all flow into the same OneVision-Encoder under a shared (t, h, w) position scheme. No task-specific tokenizers, no per-modality routing.

Four-Stage Training Curriculum

We train LLaVA-OneVision-2 in four compact stages:

1. Bootstrap video ability from LLaVA-OneVision-1.5 with short 30s video captions.
2. Instruction tune with large multimodal instruction data and 30–180s video captions.
3. Extend to long videos with 10–15 min captions and public video instruction data.
4. Refine codec, spatial, and tracking skills with denser long-video sampling, point tracking, and 4M spatial samples.

The curriculum mixes LLaVA-OneVision-1.5 data, FineVision, and new in-house video caption/spatial datasets; we do not synthesize any video instruction data.

Models

Datasets

Quick Start (4B, single node)

🚨 Strongly recommended: don’t read the rest of this section. Just hand the prompt below to any code agent (Claude Code, opencode, Cursor, Codex, etc.) from a working directory on your 8×GPU node, and it will reproduce the full SFT run end-to-end — clone, download the 340 GB bundle, load the docker image, and launch detached training — without you reading a single step.

Follow these instructions to start training: https://github.com/EvolvingLMMs-Lab/LLaVA-OneVision-2#quick-start-4b-single-node

Prefer to do it by hand? Keep reading.

Audience: code agent or human reproducing LLaVA-OneVision-2-4B SFT on a single 8×GPU node. Target script: examples/llava_onevision2/quick_start_4b/quick_start.sh Bundle: lmms-lab-encoder/ov2_quickstart on Hugging Face — contains the prebuilt docker image, the Megatron tp1pp1 checkpoint, the HF tokenizer/config, and the packed Energon dataset (~340 GB total, mostly the packed dataset).

The numbered sections below (§0 Prerequisites through §5 Run the quickstart training) are the human walkthrough — same steps, broken out with explanations and alternatives (Option B paths, interactive launch, hyperparameter overrides).

0. Prerequisites

8 × A800 (or equivalent), Docker with NVIDIA Container Toolkit.

1. Layout

Repo is mounted at /workspace/LLaVA-OneVision-2 inside the container. The HF bundle goes into /ov2_quickstart/. Launch all commands from the repo root — the training script uses relative paths.

2. Get the dataset + checkpoint

The dataset (packed_mixed_sft_cap_v30s/, ~308 GB) must come from the HF bundle — it’s pre-packed Energon shards specific to this recipe. The checkpoint has two paths: download the ready-to-train Megatron tp1pp1 (recommended) or build it yourself from the standalone ViT + LLM.

Option A — Download the full bundle (recommended)

cd 
huggingface-cli download --repo-type dataset --resume-download \
    --local-dir ./ov2_quickstart \
    lmms-lab-encoder/ov2_quickstart

~340 GB. Resumable.

Option B — Merge the checkpoint yourself

3. Get the docker image

Option A — Load the prebuilt image (recommended)

docker load -i ov2_quickstart/llava_megatron.26.05.tar
docker images | grep llava_megatron
# llava_megatron   26.05      ...   ~30GB

~1 minute. Skips the long base-image pull + dependency install.

Option B — Build from source

4. Launch the container

Use the repo’s wrapper. It mounts the host repo at /workspace/LLaVA-OneVision-2, opens all GPUs, sets NCCL env vars for IB, and drops you into a bash shell.

cd 
bash run_docker_local.sh

You should now be inside the container at:

root@:/workspace/LLaVA-OneVision-2#

5. Run the quickstart training

Inside the container, from /workspace/LLaVA-OneVision-2:

# Optional: pick an output dir on a disk with ≥ 100 GB free for checkpoints + tensorboard.
export OUTPUT_DIR=/workspace/LLaVA-OneVision-2/output/quick_start_4b

bash examples/llava_onevision2/quick_start_4b/quick_start.sh

That’s it. Defaults: 8 GPUs, TP=1, PP=1, SEQ_LEN=10192, MBS=1, GBS=16, 1 epoch over 219,907 packed bins. Logs + checkpoints land under ${OUTPUT_DIR}/quick_start/.

Contributors

Thanks so much to all of our amazing contributors!

Explore all contributors →

Citation

If you find LLaVA-OneVision-2 useful in your research, please cite the following paper:

@misc{an2026llavaonevision2nextgenerationperceptualintelligence,
  title         = {LLaVA-OneVision-2: Towards Next-Generation Perceptual Intelligence},
  author        = {Xiang An and Yin Xie and Feilong Tang and Yunyao Yan and Huajie Tan and Didi Zhu and Changrui Chen and Xiuwei Zhao and Bin Qin and Kaicheng Yang and Yifei Shen and Yuanhan Zhang and Kaichen Zhang and Wenkang Zhang and Zheng Cheng and Nansen Zhang and Chunsheng Wu and Chunjiang Ge and Zimin Ran and Dehua Song and Chunyuan Li and Shikun Feng and Ming Hu and Zhangquan Chen and Junbo Niu and Bo Li and Ziyong Feng and Ziwei Liu and Zongyuan Ge and Jiankang Deng},
  year          = {2026},
  eprint        = {2605.25979},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2605.25979}
}

@misc{an2025llavaonevision15fullyopenframework,
  title         = {LLaVA-OneVision-1.5: Fully Open Framework for Democratized Multimodal Training},
  author        = {Xiang An and Yin Xie and Kaicheng Yang and Wenkang Zhang and Xiuwei Zhao and Zheng Cheng and Yirui Wang and Songcen Xu and Changrui Chen and Didi Zhu and Chunsheng Wu and Huajie Tan and Chunyuan Li and Jing Yang and Jie Yu and Xiyao Wang and Bin Qin and Yumeng Wang and Zizhen Yan and Ziyong Feng and Ziwei Liu and Bo Li and Jiankang Deng},
  year          = {2025},
  eprint        = {2509.23661},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2509.23661}
}

@misc{tang2026onevisionencodercodecalignedsparsityfoundational,
  title         = {OneVision-Encoder: Codec-Aligned Sparsity as a Foundational Principle for Multimodal Intelligence},
  author        = {Feilong Tang and Xiang An and Yunyao Yan and Yin Xie and Bin Qin and Kaicheng Yang and Yifei Shen and Yuanhan Zhang and Chunyuan Li and Shikun Feng and Changrui Chen and Huajie Tan and Ming Hu and Manyuan Zhang and Bo Li and Ziyong Feng and Ziwei Liu and Zongyuan Ge and Jiankang Deng},
  year          = {2026},
  eprint        = {2602.08683},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2602.08683}
}

@article{li2025llavaonevisioneasyvisualtasktransfer,
  title         = {LLaVA-OneVision: Easy Visual Task Transfer},
  author        = {Bo Li and Yuanhan Zhang and Dong Guo and Renrui Zhang and Feng Li and Hao Zhang and Kaichen Zhang and Peiyuan Zhang and Yanwei Li and Ziwei Liu and Chunyuan Li},
  journal       = {Transactions on Machine Learning Research},
  year          = {2025},
  url           = {https://openreview.net/forum?id=zKv8qULV6n}
}

Acknowledgement

We extend our sincere gratitude to the LoongForge team from Baidu AI Cloud for providing the exceptional AIAK-based training framework. The outstanding capabilities of AIAK-Training-LLM and AIAK-Megatron have significantly accelerated our training process with remarkable efficiency. We are especially grateful for their citation of LLaVA-OneVision-2 and deeply appreciate their recognition and support.

We acknowledge the support of Synvo AI for contributing to the partial data annotation in this work, and also thank the maintainers and contributors of the following open-source projects, whose work greatly inspired and supported our research:

• LLaVA: Large Language-and-Vision Assistant — LLaVA
• LLaVA-NeXT: Next-generation multi-modal assistant — LLaVA-NeXT
• lmms-eval: A standardized evaluation framework for Large Multimodal Models — lmms-eval
• Megatron-LM: Efficient, scalable training for large language models — Megatron-LM
• Qwen2.5-VL: Strong vision-language foundation model — Qwen2.5-VL
• InternVL: Open-source large-scale vision-language foundation model — InternVL
• Qwen3: Next-generation Qwen LLM — Qwen
• MetaCLIP: Scalable contrastive pretraining — MetaCLIP
• FineVision: Open Data Is All You Need — FineVision