--- title: humanomniv2 canonical_url: "https://www.modelscope.cn/models/iic/humanomniv2" md_url: "https://www.modelscope.cn/models/iic/humanomniv2.md" repository: iic/humanomniv2 last_updated: 2025-06-30 license: CC-BY-NC-ND pipeline_tag: image-text-to-text tasks: - image-text-to-text model_type: - qwen2_5_omni_thinker architectures: - Qwen2_5OmniThinkerForConditionalGeneration parameters: 8.9B tensor_type: - BF16 library_name: - safetensors - pytorch frameworks: - Pytorch downloads: 583 stars: 5 --- # humanomniv2 > humanomniv2 - iic 在 ModelScope 开源的模型。HumanOmniV2: From Understanding to Omni-Modal Reasoning with Context Paper: Arxiv iic/humanomniv2 是 ModelScope 魔搭社区上的 8.9B 参数image-text-to-text模型，采用 CC-BY-NC-ND 许可。 - **Repository**: iic/humanomniv2 - **License**: CC-BY-NC-ND - **Tasks**: image-text-to-text - **Parameters**: 8.9B - **Downloads**: 583 - **Stars**: 5 - **Last updated**: 2025-06-30 Source: https://www.modelscope.cn/models/iic/humanomniv2 --- # HumanOmniV2: From Understanding to Omni-Modal Reasoning with Context - Paper: [Arxiv](https://arxiv.org/abs/2506.21277) With the rapid evolution of multimodal large language models, the capacity to deeply understand and interpret human intentions has emerged as a critical capability, which demands detailed and thoughtful reasoning. In recent studies, Reinforcement Learning (RL) has demonstrated potential in enhancing the reasoning capabilities of Large Language Models (LLMs). Nonetheless, the challenges associated with adapting RL to multimodal data and formats remain largely unaddressed. In this paper, we identify two issues in existing multimodal reasoning models: insufficient global context understanding and shortcut problems. To tackle these issues, we emphasize the necessity for the model to reason with a clear understanding of the global context within multimodal inputs. This global context understanding can effectively prevent the model from overlooking key multimodal cues and ensure a thorough reasoning process. To ensure the accurate interpretation of multimodal context information and improve complex reasoning capability, we implement context reward and logical reward judged by a large language model, alongside format and accuracy rewards. Our proposed method demonstrates advanced performance across multiple omni-modal benchmarks compared to other open-source omni-modal models. #### 🌟 Contributions in HumanOmniV2 1. We propose that models should summarize the context of multimodal inputs before engaging in the reasoning process. This approach aims to mitigate issues such as skipping crucial multimodal information and context understanding on multimodal inputs. 2. We curate a human-centric benchmark, IntentBench, for omni-modal evaluation, which requires simultaneously understanding video and audio, the global context, complex social relationships, and careful observation. 3. Our proposed HumanOmniV2 achieves the best performance across multiple omni-modal benchmarks compared to existing open-source omni-modal methods. ## 📜 License - Our models and code are under the Apache License 2.0. - But our self-collected videos are under [**CC BY-NC-SA 4.0**](https://creativecommons.org/licenses/by-nc-nd/4.0/) license. #### 您可以通过如下git clone命令，或者ModelScope SDK来下载模型 SDK下载 ```bash #安装ModelScope pip install modelscope ``` ```python #SDK模型下载 from modelscope import snapshot_download model_dir = snapshot_download('iic/humanomniv2') ``` Git下载 ``` #Git模型下载 git clone https://www.modelscope.cn/iic/humanomniv2.git ```

如果您是本模型的贡献者，我们邀请您根据模型贡献文档，及时完善模型卡片内容。