Huawei: there will be no such names as "Pangu Chat", nor will it "align" industry products.

Thanks to CTOnews.com netizen West Window old clues delivery! CTOnews.com June 5 news, the market has news that Huawei large model "Pangu Chat" has applied for relevant trademarks, will be released on July 7. According to the Securities Times, Huawei officials said Huawei would not have such names as "Pangu Chat," nor would it "benchmark" industry products, although the official calibre did not directly mention the release of Pangu's big model.

CTOnews.com inquired about public information and learned that Huawei Pangu series basic large models were officially released in 2021, including NLP (natural language processing), CV (machine vision) and scientific calculation large models; subsequently, large models of mines, drug molecules, meteorology, waves and other industries were released.

According to introduction, NLP is the first Chinese pre-trained large model with 100 billion parameters, while CV large model reaches 3 billion parameters for the first time. Pangu CV large model is the largest CV large model in the industry, achieving both discrimination and generation capabilities for the first time, and the first in the industry in terms of small sample learning capabilities on ImageNet; Pangu Meteorological Large Model provides second-level weather forecasts; Zidong. Tai Chu is the world's first three-mode model of graphics, text and sound.

For Pangu model positioning, Huawei's internal team has established three key core design principles: one is that the model should be large and can absorb massive data; the other is that the network structure should be strong and can really give full play to the performance of the model; and the third is to have excellent generalization ability, which can really land in all walks of life.

Huawei Cloud official website shows that Pangu large model consists of NLP large model, CV large model, multimodal large model, scientific calculation large model and other large models. Through model generalization, it solves AI scale and industrialization problems that cannot be solved under traditional AI workshop development mode, and can support a variety of natural language processing tasks, including text generation, text classification, question answering system, etc.

Huawei said that Pangu NLP large model was jointly developed by Huawei Cloud, Circulation Intelligence and Pengcheng Laboratory, and has leading language understanding and model generation capability: in the authoritative Chinese language understanding evaluation benchmark CLUE list, Pangu NLP large model ranked first in the overall ranking list, classification and reading comprehension items, refreshing the world history record of the three lists; The overall ranking score is 83.046, leading the industry in multiple sub-task scores, a big step towards the human level (85.61).

Specifically, Pangu NLP large model uses Encoder-Decoder architecture for the first time, which takes into account the understanding ability and generation ability of NLP large model, ensuring the embedding flexibility of model in different systems. In downstream applications, only a small number of samples and learnable parameters are needed to complete the rapid fine-tuning and downstream adaptation of the 100 billion-scale large-scale model, which has good performance in intelligent public opinion and intelligent marketing.