At the 2024 Global Developer Pioneer Conference,MoDa CommunityLaunched "ModelScope-Sora Open SourceThe project aims to promote the development of Chinese Sora Modelexploration and innovation.
The program provides a one-stop tool chain, including data processing tools, multimodal datasets, Sora-like basic models, training and inference tools, etc.
MoDa has released Data-Juicer, a multimodal data processing system that contains more than 100 efficient operators, which can greatly improve the efficiency and quality of video data processing. Data-Juicer supports text, image, audio, and video processing, and developers can freely combine operators, such as editing videos, enhancing resolution, etc.
In addition, MoDa has also launched a basic Sora-like model, lite-Sora, and will hold a "ModelScope-Sora Challenge" to encourage developers to participate in the development of Sora-like models.
In the future, MoDa plans to build an open high-quality multimodal dataset in Chinese to help the development of China's multimodal big models.
- Data-Juicer page: https://github.com/modelscope/data-juicer
- lite-Sora page: https://github.com/modelscope/lite-sora