Baichuan AI (百川智能), an innovative AI company specializing in large language models (LLM), yesterday open-sourced its LLMs for commercial use and free of charge, marking a milestone for the young company to develop China’s answer to OpenAI’s ChatGPT and Meta’s LLaMA.
Baichuan AI unveiled its Baichuan2-7B, Baichuan2-13B、Baichuan2-13B-Chat and its 4bit edition, after fine-tuning and open-sourcing them.
Notably, Wang Xiaochuan, founder of Baichuan AI, claimed its performance “far outstrips” that of LLaMA2.
The former founder and CEO of Sogou (搜狗), a search engine service provider, said Baichuan7B features 7 billion parameters.
This already rivals the English-language grading of LLaMA2, which is built on the basis of 13 billion parameters, he noted.
“As such, we can say the smaller LLM is on par with the bigger counterpart,” he said at the launch event. “This means our bigger LLM has better performance and is ahead of LLaMA2 in every aspect.”
The new releases came as a number of Chinese AI firms have been approved to offer chatbot services based on generative AI and LLM technologies.
Since its founding in April this year, Baichuan AI has made significant inroads in its quest to roll out an equivalent of ChatGPT.
To date, downloads of its Baichuan-7B and Baichuan-13B models have topped 5 million.
The LLMs it introduced this time around has met with positive feedback from industrial chain partners.
Established corporations like Tencent Cloud, AliCloud, Volcengine, Huawei and MediaTek have attended the launch event by Baichuan AI and forged partnerships with the startup.