} ?>
(Yicai) April 23 -- Llama 3, the latest series of Meta's open-source large language models, will be accessible to developers through major Chinese cloud providers Baidu, Alibaba Group Holding, and Tencent Holdings.
The three Chinese tech giants rushed to support the newest rival of OpenAI's ChatGPT, released by the owner of Facebook on April 18.
Baidu AI Cloud said on April 19 that it has become the first cloud service provider in China to launch training and inference solutions for the full range of Llama 3 models.
Alibaba Cloud announced yesterday that for a limited time, it will offer free Llama 3 training and inference. Moreover, Tencent Cloud said that it has become one of the first domestic platforms to support all Llama 3 models.
The three internet giants are working on their own LLMs. Tencent revealed Tencent Hunyuan last September while Alibaba opened access to Tongyi Qianwen to the public that month. Baidu unveiled its first Ernie Bot in March 2023. Besides these homegrown innovations, the firms want to get a piece of the computing power demand arising from the use of other LLMs.
Zhou Hongyi, founder and chairman of Qihoo 360 Technology, said that the performance of Llama 3 is strong, and might even surpass GPT-3.5, a model released in March 2022. The latest ChatGPT version is GPT-4 Turbo, launched last November.
But Meta is determined to win the competition with OpenAI as several tech industry insiders said that if Meta releases the most powerful Llama 3 version that it is training with over 400 billion parameters it could reshape the landscape of LLMs.
However, industry insiders pointed out that Llama 3 is not very strong in using Chinese or handling long texts. A developer of LLMs conducted a test and said that the Chinese language proficiency of Llama 3 is not as good as that of GPT-3.5.
Some analysts pointed out that the latest Llama series is weaker in Chinese but this may not be such a big problem because developers do not necessarily need vast amounts of domestic data to make great models.
Editor: Emmi Laine