Alibaba announced the launch of the "Tongyi" big model, and the core model has been opened up to developers around the world.

CTOnews.com Sept. 2, today, at the theme forum of the World artificial Intelligence Congress, Alibaba Dharma House announced the release of the latest "Tongyi" large model series, saying that the relevant core models are open to developers around the world.

In addition, in order to speed up the large-scale application of the large model, the Dharma Institute has also developed the S4 framework, the key technology for the landing of the super-large model. The multi-task accuracy of the large-scale model with 10 billion parameters can be close to lossless when the compression ratio reaches 99%.

According to reports, through the deployment of super-large models and lightweight versions, Alibaba Tongyi large model series has provided services in more than 200 scenarios, achieving a 10% improvement in application effect by 2%. Typical usage scenarios include e-commerce cross-modal search, AI-aided design, open domain man-machine dialogue, legal document learning, medical text understanding and so on.

CTOnews.com learned that, at the same time as this release, core models and capabilities such as language large model AliceMind-PLUG, multi-modal understanding and generation unified model AliceMind-mPLUG, multi-modal unified base model M6-OFA, and S4 framework, key technology for super-large model landing in Tongyi large model series have been opened to developers around the world.