Alibaba Releases Qwen2-VL AI Model with Vision-Language Capabilities, 2B and 7B Sizes, Apache 2.0 License

Aug 29, 2024, 04:14 PM

Alibaba has announced the release of its new AI model, Qwen2-VL, which includes vision-language models with state-of-the-art capabilities. The models, available in 2B and 7B sizes, are open-sourced under the Apache 2.0 license. They excel in visual understanding and video analysis, capable of comprehending videos over 20 minutes long. The Qwen2-VL models feature Naive Dynamic Resolution and multimodal RoPE, enhancing their ability to handle various image resolutions and ratios. They also offer multilingual support, including for CJK, Arabic, and European languages, and can be integrated into mobile and robotic applications. The largest model, Qwen2-VL-72B, is available via API and demonstrates superior performance compared to previous models.