Alibaba Cloud launches open source Large Vision Language Model Qwen-VL

On August 25, Alibaba Cloud launched an open-source Large Vision Language Model (LVLM) named Qwen-VL. The LVLM is predicated on Alibaba Cloud’s 7 billion parameter foundational language mannequin Qwen-7B. In addition to capabilities reminiscent of image-text recognition, description, and query answering, Qwen-VL introduces new options together with visible location recognition and image-text comprehension, the corporate mentioned in an announcement. These capabilities allow the mannequin to establish places in footage and to offer customers with steering primarily based on the data extracted from pictures, the agency added. The mannequin will be utilized in varied situations together with picture and document-based query answering, picture caption era, and fine-grained visible recognition. Currently, each Qwen-VL and its visible AI assistant Qwen-VL-Chat can be found free of charge and business use on Alibaba’s “Model as a Service” platform ModelScope. [Alibaba Cloud statement, in Chinese]