Alibaba Cloud has introduced two open-source large vision language models (LVLM): Qwen-VL and Qwen-VL-Chat. These models can understand images, texts, and prompts, enabling multi-round question answering in English and Chinese. Alibaba Cloud aims to democratize AI technology by sharing the models with the open-source community and commercial institutions.
Facts
- Alibaba Cloud launches open-source vision language models (LVLM), Qwen-VL and Qwen-VL-Chat.
- Qwen-VL is a multimodal model capable of understanding both image inputs and text prompts in English and Chinese, performing tasks like open-ended queries and generating image captions.
- Qwen-VL-Chat enables complex interactions, such as comparing multiple images and engaging in multi-round question answering, showcasing creative capabilities.
- Alibaba Cloud has shared the model’s code, weights, and documentation with academics, researchers, and commercial institutions worldwide to democratize AI technologies.
- Models are accessible via Alibaba’s AI model community ModelScope and Hugging Face for commercial use.
- These models have the potential to revolutionize interactions with visual content, aiding visually impaired individuals during online shopping.
- Qwen-VL handles image input at a resolution of 448×448, resulting in better image recognition and comprehension.
- Qwen-VL recorded outstanding performances on various visual language tasks and benchmarks.
- Qwen-VL-Chat achieved leading results in text-image dialogue and alignment with humans.
- Alibaba Cloud previously open-sourced Qwen-7B and Qwen-7B-Chat, two 7-billion-parameter LLMs, with over 400,000 downloads within a month of their launch.
