Shulou(Shulou.com)11/24 Report--

[Beijing, September 6, 2023] today, Baichuan Intelligent held a large model conference in Beijing, officially released the Baichuan2 open source big model, Teng AI basic software and hardware platform officially supports the Baichuan2 big model, and launched the Baichuan2-7B model open experience on the open source MindSpore open source community big model platform.

At the press conference, Baichuan Intelligence announced that Baichuan2-7B, Baichuan2-13B and Baichuan2-13B-Chat, together with their quantitative versions of 4bit, will provide large model services for academic and commercial markets, and they are all free and commercially available.

Excellent performance in both arts and science, leading LLaMA 2 in an all-round way

Both Baichuan2-7B-Base and Baichuan2-13B-Base are based on 2.6T "quality multilingual data entry training". While retaining the good generation and creation ability of the previous generation open source model, fluent multi-round dialogue ability and low deployment barriers, the two models have significantly improved in mathematics, code, security, logical reasoning, semantic understanding and so on. Compared with the previous generation 13B model, Baichuan2-13B-Base can improve math ability by 49%, code by 46%, security ability by 37%, logical reasoning ability by 25%, and semantic understanding ability by 15%.

The performance of the two open source models is excellent in each evaluation list, leading LLaMA 2 by absolute advantage in several authoritative evaluation benchmarks such as MMLU, CMMLU, GSM8K, etc., compared with other models with the same number of parameters, the performance is also very outstanding, and the performance is significantly better than that of LLaMA 2 models of the same size.

What is more worth mentioning is that according to MMLU and other authoritative English evaluation benchmark scores of Baichuan2-7b, with a parameter of 7 billion, it is basically the same as the 13 billion parameters of LLaMA 2 in English mainstream tasks.

Benchmark score of 7B parameter model

Benchmark score of 13B parameter model

Baichuan2-7B and Baichuan2-13B are not only completely open to academic research, developers can also apply for an official commercial license by email.

Baichuan2 large model

The Baichuan2 model is a series of open source and commercially available large-scale pre-training language models developed by Baichuan Intelligence. The model contains 7 billion, 13 billion, 53 billion parameters. At the beginning of Baichuan Intelligent's success, it will be an important development direction of the company to help the prosperity of China's "model" through open source. The two open source Baichuan2 models received a positive response from upstream and downstream enterprises. Huawei and many other well-known enterprises participated in this conference and reached cooperation with Baichuan Intelligence.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.