On September 7, at the 2023 Tencent Global Digital Ecology Conference, Tencent's self-developed universal big language model, the hybrid model, was officially unveiled.。

According to the official introduction, Tencent's hybrid model, by Tencent from the first token from zero training, inside the algorithm, framework, platform full link self-research.。In the hybrid model, Tencent has developed its own machine learning framework Angel, which doubles the training speed compared to mainstream frameworks in the industry and increases the inference speed by 1.3 times。The hybrid model has a parameter scale of over 100 billion yuan and a pre-training corpus of over 2 trillion tokens, which makes the model have strong Chinese writing ability, logical reasoning ability in complex contexts, and reliable task execution ability.。

Specifically, Tencent has optimized the pre-training algorithm and strategy to reduce the illusion of the hybrid model by 30% to 50% compared to the mainstream open source model, which can reduce the "nonsense" when answering.。Secondly, through the reinforcement learning method, let the model learn to identify trap problems, and now the rejection rate of the hybrid model in the face of security-induced class problems has increased by 20%.。Furthermore, Tencent has improved the processing effect and performance of the hybrid model for ultra-long texts through position coding optimization, which can generate thousands of words in one breath.。In addition, the hybrid model has stronger logical reasoning ability, and can make reasoning and decision-making in combination with actual application scenarios like people.。

Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of Cloud and Smart Industry Group, said: "With big model generation technology at its core, artificial intelligence is becoming a key driver of the next round of digital development.。"

At present, more than 50 Tencent businesses and products, including Tencent Cloud, Tencent Advertising, Tencent Games, Tencent Fintech, Tencent Meetings, Tencent Docs, WeChat Search and QQ Browser, have been connected to the Tencent Hybrid Model Test and have achieved initial results.。

In addition to disclosing the hybrid model, Tencent also announced that the hybrid model is officially open to the public through Tencent Cloud.。Users can call the experience directly on Tengxun Cloud through the API, or use the hybrid as the base model to fine tune on the public cloud.。

At present, the WeChat applet has launched the internal beta version of the hybrid model, the use of the internal beta version needs to apply for an appointment, after the appointment to wait in line for the trial.。

It has become a trend for big models to open up to the whole society, and a "big war" has begun.

Before Tencent, a number of "friends" have opened up their own large models, such as Baidu's "Wen Xin Yi Yan," Shang Tang technology model application "to discuss SenseChat," Baichuan intelligent "Baichuan model," HKUST's "Xunfei spark" cognitive model and so on.。

Among them, Baidu Wenxin Yiyan opened its website and mobile App to the whole society on August 31.。On the same day, Wenxin Yiyan APP topped the Apple Store free list, becoming the first Chinese AI native app to top the App Store list.。

As one of the first AI companies in China to deploy the AIDC Artificial Intelligence Supercomputing Center, Shang Tang Technology is not far behind.。Shang Tang Technology officially opened to users on August 31 to discuss SenseChat。

On August 31, Baichuan Intelligence announced that its "Baichuan Model" was filed through the Interim Measures for the Management of Generative Artificial Intelligence Services and opened to the whole society.。Unlike Baidu, Shang Tang Technology and other technology companies deep AI for many years, Baichuan Intelligence was established in April this year.。It is understood that the company was founded by former Sogou CEO Wang Xiaochuan, the team has a number of well-known "big factory" AI top talent from Sogou, Baidu, Huawei, ByteDance, Tencent, etc.。And just two months after its establishment, Baichuan Intelligence has released three universal large language models.。Baichuan Intelligence released the first open source and free commercial large language model Baichuan-7B in June this year, the large language model Baichuan-13B-Base and the dialogue model Baichuan-13B-Chat with 13 billion parameters in July, and the large language model Baichuan-53B with 53 billion parameters in August.。

Then, the "national team" iFLYTEK also announced on September 5 that the iFLYTEK Spark cognitive model is open to the whole people.。Like Baidu's Wenxin, iFLYTEK has also opened two channels for "iFLYTEK Spark" to register on the official website and download apps from the app store.。On May 6 this year, iFLYTEK officially released the "iFLYTEK Spark" cognitive model, which was upgraded to iFLYTEK Spark V1 in June and August..5. Xunfei Spark V2.Version 0。It is reported that in the August 17 "MIT Science and Technology Review" China released a large model evaluation report, Xunfei Spark V2.0 with a total score of 81.5 points at the top of the list。

In addition to the aforementioned technology companies, large models of companies and institutions such as Huawei, ByteDance, Alibaba, Zidong Taichu, and Zhipu Huazhang are also on their way.。Judging from the big models released so far, opening up to the whole society has become a trend.。Follow-up other enterprises on-line big model, the probability will also take the "full opening" this road.。

Many technology companies choose to open up to the whole society, which is actually a "win-win" business.。For the general public, you can use a number of AI models, which can improve the efficiency of learning and work, and embrace advanced artificial intelligence technology。For enterprises, opening up large models allows them to get a lot of real-world human feedback, which helps them to further improve the underlying model, continuously update iterative large model versions, and create a better user experience.。

And with more and more big models open to society, technology companies will inevitably face increasingly fierce competition.。As a "high" technology-content product, the big model still needs to win with technology in the end.。It can be predicted that a big AI model of survival of the fittest will be staged in the near future。