On May 24, Tencent’s “Hunyuan” AI model reached the top of the CLUE (Chinese Language Understanding Evaluation Collection) overall ranking, reading comprehension, and large-scale knowledge graph at the same time, breaking three records in one fell swoop.
It is understood that the CLUE total list consists of classification tasks and reading comprehension tasks. Tencent’s “Hunyuan” AI model achieved double success in classification tasks and reading comprehension within a month, and finally ranked first in the overall list with a score of 84.730.
As one of the most authoritative natural language understanding lists in Chinese, CLUE has set up 9 sub-tasks including text similarity, classification, contextual reasoning, and reading comprehension, aiming to promote NLP (Natural Language Processing Pre-training) models The continuous progress and breakthrough of technology.
NLP (Natural Language Processing) technology is a core research direction in the field of artificial intelligence. Its purpose is to enable computers to have human listening, speaking, reading, and writing capabilities, and to use knowledge and common sense for reasoning and decision-making. At present, more and more technology companies and R&D institutions are investing in research in this field, and the competition in industry rankings such as CLUE is also fierce.
In the pre-training stage, in addition to the regular public datasets, the “Hunyuan” large model also learns the text datasets specific to the business domain. Therefore, compared with other large AI models in the industry, “Hunyuan” can better understand text information of various lengths, and can cope with diverse scene tasks such as search, advertising, news, and question and answer. The task is also more advantageous.
In addition to performance improvement, the “Hunyuan” large model effectively compresses the amount of communication data trained by GPU nodes in low-bandwidth environments through methods such as data and model course learning, multi-sentence merging masks, and improved PowerSGD (optimized communication algorithm). And communication is time-consuming, and the training efficiency is greatly improved.
At present, Tencent’s “Hunyuan” AI large model research and development team has contributed the improved PowerSGD method to the PyTorch open source community, which will be officially launched in the next version of PyTorch.
Thanks to the powerful technical capabilities of the “Hunyuan” AI model in the fields of natural language understanding and cross-modal retrieval, since April this year, the model has successfully won major authoritative AI lists such as MSR-VTT, MSVD, and CLUE. Top of the list, which means that Tencent has made breakthroughs in technological research and development in the field of artificial intelligence.
At present, the “Hunyuan” NLP model has been applied to multiple businesses within Tencent, and has brought more than 5% index improvement in Tencent’s advertising data mining tasks, improving the accuracy of advertising recommendations and optimizing user experience. In the future, Tencent Hunyuan AI large model research and development team will continue to promote the research and optimization of large models according to the needs of specific scenarios, and accelerate the application and implementation of AI technology in various industries.
Leifeng Network
This article is reproduced from: https://www.leiphone.com/category/industrynews/34hFPKlJeZRQD73A.html
This site is for inclusion only, and the copyright belongs to the original author.