According to public information, the AI Large-Scale Model Technology Summit hosted by the Chinese Association for Artificial Intelligence will be held in Hangzhou from April 8 to 9. The conference agenda shows that Huawei Cloud’s Chief Scientist in Artificial Intelligence, Qi Tian, will introduce “The Progress and Application of Pangu Large-Scale Models” at the event.
Huawei has been investing in large-scale models since 2020. The Pangu series consists of CV (computer vision) large-scale models, scientific computing big models, NLP (natural language processing) large-scale models, voice recognition large-scale models and multimodal large-scale models. It was officially released in April 2021. At that time, Huawei stated that Pangu was the industry’s first hundred-billion-level NLP model for generating and understanding Chinese.
According to Huawei, during the pre-training phase of its NLP Large-Scale Model, the model learned from a total of 40TB of Chinese text data, making it considered an AI model with the closest ability to human Chinese comprehension. The CV Large-Scale Model is currently regarded as having both image discrimination and generation capabilities at once along with being able to extract according to demand, ranking first among small sample learning abilities on ImageNet. The Scientific Computing Large-Scale Model is applied in fields such as meteorology and biomedicine.
A research report by CITIC Securities pointed out that the current capability of Pangu has already been verified in areas including energy consumption, the retail finance industry, and others. Meanwhile, Pangu’s goal for few-shot learning through transfer learning has incorporated latest technologies like P-tuning into Pangu’s fine-tuning framework and trained it through a huge industry knowledge base, making it more suitable for complex commercial scenarios than GPT-3.
It is worth mentioning that at the Huawei New Product Launch on March 23, Huawei brought a new smart search function based on multi-modal large-scale model technology. The model was miniaturized for smartphone users, which also made Huawei’s Pangu Large-Scale Models become the focus of attention of various market institutions.
The launch of OpenAI’s GPT large-scale model has ushered in a new era of general AI, leading many companies to develop their own AI models. Among the international players are Google and Meta, while Chinese companies such as Huawei, Baidu, Alibaba, and Tencent are also competing vigorously in this field