Tencent's self-developed Hunyuan large model is officially unveiled and is open to the outside world through Tencent Cloud

Domestic large models have entered a long-term running period, shifting from parameter first to practical priority.

On September 7, at the 2023 Tencent Global Digital Ecology Conference, 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 universal large language model self-developed by Tencent Full Link. It has a parameter scale of over 100 billion and a pre-training corpus of over 20,000 Yitokens has strong Chinese creation ability, logical reasoning ability in complex contexts, and reliable task execution ability.

It is worth noting that Tencent’s Hunyuan model is a practical-level 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. .

It is understood that the Hunyuan large model will serve as the base of Tencent Cloud MaaS service. Customers can not only call Hunyuan directly through the API, but also use Hunyuan as Base model to build exclusive applications for different industrial scenarios.

Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of Cloud and Smart Industry Group, said: "With large model generation technology as the core, artificial intelligence is becoming the next The key driving force for a round of digital development also brings new ideas to solve industry pain points. Large models need to be based on industrial scenarios and integrated with enterprise data to release the greatest value."

Jiuweigong, self-developed full-link technology

According to Jiang Jie, Vice President of Tencent Group , Tencent Hunyuan large model was trained from scratch from the first token, and it mastered the full-link self-developed technology from model algorithm to machine learning framework to AI infrastructure.

Tencent Group Vice President Jiang Jie

Starting from 2021, Tencent has successively launched large sparse NLP models with hundreds of billions and trillions of parameters, breaking the three major CLUE ranking records and achieving new breakthroughs in Chinese understanding capabilities.

#At present, the application of large models in the industry is still limited, mainly focusing on leisure scenes with high fault tolerance and simple tasks. Tencent has conducted a series of self-research innovations at the algorithm level to improve model reliability and maturity.

In response to the problem that large models are prone to "gibberish", Tencent has optimized the pre-training algorithm and strategy to make the illusion of Hunyuan large models better than mainstream open source The large model has been reduced by 30% to 50%; through reinforcement learning methods, the model can learn to identify trap problems; through position coding optimization, the processing effect and performance of very long documents have been improved; a new strategy of thinking chain has been proposed to allow large models to Reason and make decisions based on actual application scenarios like a human being.

In addition, Tencent has also developed its own machine learning framework Angel, which doubles the training speed and 1.3 times the inference speed compared to the industry’s mainstream frameworks. times.

Tencent Hunyuan large model full-link self-research

got Thanks to the full-link self-developed technology, Tencent's Hunyuan large model can understand the meaning of the context and has the ability to memorize long texts, allowing it to smoothly conduct multiple rounds of conversations in professional fields. In addition, it can also create content such as literary creation, text summaries, and role plays to fully understand user intentions and provide timely responses efficiently and accurately.

In the standard compliance test of the China Academy of Information and Communications Technology's "Evaluation Methods for Large-scale Pre-training Model Technology and Applications", a total of 66 Hunyuan large models were evaluated In terms of ability items, the comprehensive evaluation in the two important areas of "model development" and "model capability" has obtained the current highest score. In the mainstream evaluation sets MMLU, CEval and AGI-eval, the Hunyuan large model has excellent performance, especially in Chinese science, college entrance examination questions, mathematics and other sub-items

.

Liu Yuanchun, President of Shanghai University of Finance and Economics, believes: “With the help of full-link self-research, China will continue to accumulate talents and technologies related to large models, and gradually form a systematic industrial chain, talent chain, technology chain and innovation chain. , and finally find a Chinese path to develop general artificial intelligence, helping us achieve breakthrough progress in digital technology innovation."

Tencent fully embraces Large model

Jiang Jie said: “Our goal in developing large models is not to obtain high scores in evaluations, but to apply the technology to actual scenarios. . Tencent will fully embrace the big model."

At this conference, Jiang Jie showed Tencent conferences, Tencent documents, Tencent advertising, etc. The actual application of multiple businesses after accessing Tencent's Hunyuan large model.

For example, Tencent Conference has created an AI assistant based on the Hunyuan large model. It only needs simple natural language instructions to complete conference information extraction, content analysis, etc. For complex tasks, intelligent summary minutes can be generated after the meeting. According to actual measurements, the Hunyuan large model has achieved a high user adoption rate in many aspects such as instruction understanding, in-meeting Q&A, meeting summaries, and meeting to-do items.

Application of Tencent Hunyuan Large Model in Tencent Conference

## In terms of document processing, Tencent Hunyuan's large model supports dozens of text creation scenarios and has been used 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. These functions are currently in the internal testing stage and will be open to users when mature.

In the advertising business scenario, Tencent’s Hunyuan model supports the creation of intelligent advertising materials, which can adapt to industry and regional characteristics and meet the needs of thousands of people. , realizing the natural integration of text, pictures and videos. In addition, based on the capabilities of the Hunyuan large model, advertising intelligent shopping guides can help merchants improve service quality and efficiency in scenarios such as corporate WeChat.

The application of Tencent Hunyuan model in Tencent advertising

## It is understood that in June this year, Tencent Cloud launched the Model as a Service (MaaS) solution, providing one-stop industry large model services covering model pre-training, model fine-tuning, and intelligent application development.

Recently, Tencent Cloud has also fully integrated into more than 20 mainstream models such as Llama 2 and Bloom. Like Hunyuan, they all support direct deployment calls. Customers can create their own exclusive industry models based on Hunyuan or open source models based on actual needs.

The above is the detailed content of Tencent's self-developed Hunyuan large model is officially unveiled and is open to the outside world through Tencent Cloud. For more information, please follow other related articles on the PHP Chinese website!