Original | Text by BFT Robot
Tencent Hunyuan large model "domineering" unveiled
On September 7, at the 2023 Tencent Global Digital Ecology Conference held in Shenzhen, Tencent's Hunyuan large model was officially unveiled and announced that it would be open to the outside world through Tencent Cloud. Tencent Hunyuan Large Model is a general-purpose large language model self-developed by Tencent Full Link. It has a parameter scale of over 100 billion and pre-training corpus of over 2 trillion tokens. It has strong Chinese creation capabilities and logical reasoning capabilities in complex contexts. , and reliable task execution capabilities.
Facing the market environment of “Battle of Hundreds of Models”, what is the key to Hunyuan’s sudden victory?
Jiang Jie, vice president of Tencent Group, said: "Tencent's strategy: first, to conquer the Chinese field, so that large models have stronger Chinese creation capabilities and improve localization capabilities when serving Chinese enterprises; second, in the complex reasoning process to enhance the ability to control the security of large models.”
It is worth noting that although the Hunyuan large model is still at a mature stage and has insufficient ability to handle complex tasks, it is a primary model that is being improved. Even though the Hunyuan model is not mature enough, Hunyuan still has a unique advantage among many foreign R&D models. This is because the Hunyuan model mainly focuses on the Chinese field and serves most domestic enterprises. Compared with many foreign English models, Domestic enterprises will choose Hunyuan for greater convenience and efficiency, as well as a sense of identification with national technology.
Hunyuan’s “draft net” has covered more than 50 industries of Tencent
Tencent's Hunyuan large model is a practical-level large model that "comes from practice and goes to practice". More than 50 Tencent businesses and products, including Tencent Cloud, Tencent Advertising, Tencent Games, Tencent Financial Technology, Tencent Conference, Tencent Documents, WeChat Souyisou, and QQ Browser, have been connected to the Tencent Hunyuan large model for testing and have achieved initial results. .
According to Tang Daosheng, vice president of Tencent Group, thousands of industries can also call Hunyuan through API, or use Hunyuan as a base model to build large model applications for different industrial scenarios. To this end, Tencent has worked closely with 11,000 ecological partners and launched industry solutions covering more than 100 industrial scenarios.
From zero to one, insist on self-research of full-link technology
It is understood that Tencent's Hunyuan large model has been trained from scratch starting from the first token, and has mastered the full-link self-developed technology from model algorithms to machine learning frameworks to AI infrastructure.
In this regard, Tencent Vice President Jiang Jie summarized the three major characteristics of the Hunyuan model: strong Chinese creation capabilities, logical reasoning capabilities in complex contexts, and reliable task execution capabilities. In the past and today, many large models have limitations in terms of performance and algorithms, and most of them are used in some simple scenes and cannot meet the adaptability to complex scenes. For example: in terms of document processing, Tencent Hunyuan large model supports dozens of text creation scenarios, and has been applied in the intelligent assistant function launched by Tencent Documents. At the same time, Hunyuan can also generate standard format text with one click, is proficient in hundreds of Excel formulas, supports natural language generation functions, and generates charts based on table content. However, chatGOT4 cannot even satisfy the document content of 4,000 words.
Jiang Jie said: "Tencent insists on self-research of technology because if a company does not do self-research from scratch, it will lack complete mastery of this technology. Tencent's self-research of large models can accelerate subsequent iterations and speed up integration with other businesses. Deep integration and binding. For Tencent’s many massive and highly concurrent businesses, the open source architecture cannot cope with the impact and is not suitable for Tencent. Therefore, we must find a research and development path based on independent systems." So far Tencent I have found a broad road suitable for my own development. In addition, Jiang Jie also said that Tencent’s self-developed machine learning framework Angel has doubled the training speed and 1.3 times the inference speed compared to the industry’s mainstream frameworks.
The launch of Tencent's Hunyuan large model has ushered in a new era of large model development in China, which is worthy of recognition and support. In addition, Tencent Cloud has fully integrated into more than 20 mainstream models such as Llama 2 and Bloom, and supports direct deployment and calling. Customers can build their own industry models based on actual needs, either based on the Hunyuan model or the open source model.
Author | Chunhua
Typography | Spring Flowers
Review | Orange Orange
If you have any questions about the content of this article, please contact us and we will respond promptly. For more information, please pay attention to BFT intelligent robot system~