“Able to write poems and write words, and be more like humans than humans!” This is the surprise brought by the world‘s largest artificial intelligence massive model-“Source 1.0” released by Inspur on September 28. In addition to being able to compose poems, it can also talk, write couplets, generate news, and continue to write stories. More importantly, it will be used in many killer scenarios such as intelligent operation and maintenance, intelligent office, and intelligent dialogue in the industry.
It is reported that the “source” single model parameters released by Inspur Artificial Intelligence Research Institute amounted to 245.7 billion, surpassing the GPT-3 developed by the OpenAI organization in the United States. The -3 model has 175 billion parameters and 570GB training data set. The source 1.0 parameter scale leads 40%, and the training data set scale leads nearly 10 times.
In terms of language intelligence, “Source 1.0” performed the best. It won the championship of the zero sample learning and small sample learning of the CLUE list of the Chinese language comprehension evaluation benchmark. In the zero-sample learning list, “Source 1.0” surpassed the industry’s best score by 18.3%, and won the championship in 6 tasks such as document classification, news classification, product classification, native Chinese reasoning, idiom reading comprehension, and noun pronoun relations; Four tasks including document classification, product classification, document summary recognition, and noun-pronoun relationship of small sample learning won the championship. In the idiom reading comprehension fill-in-the-blank project, the performance of Yuan 1.0 has surpassed the human score.
(ZeroCLUE zero sample learning list-the first human score)
Regarding the application of this model, Liu Jun, vice president of Inspur Information and general manager of AI&HPC product line, said that taking intelligent customer service as an example, when there are 10% question and answer that violates common sense, customers will have a poor experience and will not continue to talk to them. From the perspective of intelligent technology, this is because before the emergence of large models, companies used small models to apply data sets in some fields to overtrain them, and come up with a model to support the dialogue system of intelligent customer service, so it is easy to learn from common sense. Sexual problems. An important feature of the large model is that it can intelligently identify more comprehensive implicit knowledge, logic, and functions. Not only will it not make mistakes in common sense, but it can also stimulate the interaction of people to associate. This is actually Source 1.0 One of the biggest contributions.
In the future, smart operation and maintenance in operators, automatic generation of reports in smart office scenarios, smart dialogue assistants on the Internet side of mobile phones… Source 1.0 can be applied to many killer scenarios, and the prospects are broad. For Chinese academia and industry, it is possible to use a general massive language model to greatly reduce the difficulty of language model adaptation for different application scenarios, and at the same time improve the general model of small-sample learning and zero-sample learning scenarios.化 Application Ability.
It is reported that “Source 1.0” will open source, open, and share for academic research units and industrial practice users, lower the threshold for massive model research and application, accelerate the promotion of artificial intelligence research and innovation and industrial development in my country, and effectively promote the industrialization and industrialization of AI. AI turns into the fast lane.Return to Sohu to see more
Editor:
Disclaimer: The opinions of this article only represent the author himself. Sohu is an information publishing platform. Sohu only provides information storage space services.
.