1. In April 2021, Huawei Cloud released the Pangu series of ultra-large scale pre-training models, including one of the world's largest vision (CV) pre-training models with 3 billion parameters, as well as the world's largest Chinese language (NLP) pre-training model with hundreds of billions of parameters and 40 terabytes of training data, which was jointly developed with Cyclic Intelligence and Pangcheng Lab. Subsequently, Huawei Cloud will also release multi-modal, scientific computing, and other multi-disciplinary pre-training models.
2. There are three core guidelines for Pangu's design: first, an oversized neural network that can absorb massive amounts of data; second, a strong network architecture that can achieve the ultimate in performance; and third, excellent generalization capabilities that allow for full scene coverage of more than 10 times to reach all-around champion.