Baichuan 2 is an open source model released by Baichuan Intelligence, which is said to be "fully ahead of LLaMA 2 in both arts and sciences". Baichuan Intelligence officially open-sourced the fine-tuned Baichuan 2-7B, Baichuan 2-13B and Baichuan 2-13B-Chat and their 4bit quantized versions at the conference themed "Baichuan converges to the sea, open source and win-win". The models are free and commercially available.
In addition to the open source model itself, Baichuan Intelligence has also publicized Check Points in the model training process and announced that it will release a Baichuan 2 technical report detailing the training details of Baichuan 2. You can find the download address of Baichuan 2 in this link: https://github.com/baichuan-inc/Baichuan2
According to Baichuan Intelligence's official introduction, both Baichuan2-7B-Base and Baichuan2-13B-Base have been trained based on 2.6 trillion high-quality multilingual data. Both models have significantly improved their capabilities in math, code, security, logical reasoning and semantic understanding. Specifically, Baichuan2-13B-Base has 491 TP3T of improvement in math capability, 461 TP3T in code capability, 371 TP3T in security capability, 251 TP3T in logical reasoning capability, and 151 TP3T in semantic comprehension capability compared to the previous-generation 13B model.
In the benchmark tests of MMLU, CMMLU and GSM8K, both models of BCinks Intelligence have higher scores than LLaMA 2, and have a greater advantage in the "Chinese universal" part.
Baichuan2-7B and Baichuan2-13B are not only completely open to academic research, but developers can also commercialize them for free by simply applying for an official commercial license via email. Tencent Cloud, Ali Cloud, and Volcano Ark,Huawei (brand)The company, MediaTek and other companies participated in the conference and reached a cooperation with BCinks Intelligence.
This article comes from users or anonymous contributions, does not represent the position of Mass Intelligence; all content (including images, videos, etc.) in this article are copyrighted by the original author. Please refer to this site for the relevant issues involvedstatement denying or limiting responsibilityPlease contact the operator of this website for any infringement of rights (Contact Us) We will handle this as stated. Link to this article: https://dzzn.com/en/2023/825.html
