By Greg Gao
(JW Insights) Sep 7 -- China’s technology titan Tencent officially launched its in-house developed universal large language model(LLM), Hunyuan, on September 7.
This LLM boasts over a hundred billion parameters and has been trained on more than 2 trillion tokens of text data. It exhibits strong Chinese language generation capabilities, complex context logic reasoning abilities, and reliable task execution capabilities, according to Tencent.
Compared to OpenAI’s ChatGPT 3.5/4.0, Hunyuan has reduced the occurrence of fabricated information by 30%-50%, resulting in fewer irrelevant descriptions during content generation. Furthermore, when facing improper questions, it can achieve a 20% increase in refusal rate, said the company.
Hunyuan can also provide ultra-long text generation capabilities and, when combined with plugins, can generate texts as long as four thousand characters. The product showcases enhanced logical reasoning abilities and can make reasoning decisions based on real-world scenarios.
Tencent stated that its LLM has undergone internal testing in over 50 of its products and services. Hunyuan, with a strong Mandarin comprehension and creation capabilities, has already been integrated into the company’s various business lines such as WeChat, cloud, advertising, gaming, financial technology, online conferences, and documents, Tencent’s senior executive vice-president Dowson Tong, said at Tencent Global Digital Ecosystem Summit on September 6.
On the same day, Tencent announced that Hunyuan is now officially available to the public through Tencent Cloud. Users can directly access the API interface or fine-tune the LLM as a base model on the public cloud.
The move came after Beijing granted the first batch of approvals for generative AI services in August and gave the green light to technology firms, including search giant Baidu and AI firm SenseTime, to offer ChatGPT-like chatbots to the public, said China Daily.
In June, Tencent Cloud, the company’s cloud subsidiary, already launched an industry-specific large model. Compared with general LLMs like ChatGPT, industry-specific large models are basically industrial versions of ChatGPT focused on specialized niche sectors.
Tencent Cloud said that industry-specific large model solution has been applied to 10 major industries, such as finance, culture and tourism, government affairs, media, and education, and is able to offer 50 different kinds of solutions, according to China Daily.
