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

Thank you, Mr. Air, a netizen of CTOnews.com, for your clue delivery! CTOnews.com July 11 news, Wang Xiaochuan's Baichuan Intelligence today released the Baichuan-13B model, known as "13 billion parameters open source commercial."

▲ source Baichuang-13B GitHub page according to official introduction, Baichuan-13B is an open source and commercially available large-scale language model with 13 billion parameters developed by Baichuan Intelligent after Baichuan-7B. It achieves the best results in the same size model in both Chinese and English Benchmark. This release includes pre-training (Baichuan-13B-Base) and alignment (Baichuan-13B-Chat) versions.

The ▲ source Baichuang-13B GitHub page officially declares that Baichuan-13B has the following characteristics:

Larger size and more data: Baichuan-13B further expanded the number of parameters to 13 billion on the basis of Baichuan-7B, and trained 1.4 trillion tokens on high-quality corpus, surpassing LLaMA-13B40%, is currently the model with the largest amount of training data under the open source 13B size. Support for both Chinese and English, using ALiBi location coding, context window length is 4096.

At the same time, open source pre-training and alignment model: the pre-training model is a "base" for developers, while the majority of ordinary users have a stronger demand for alignment models with dialogue functions. Therefore, the project also has the alignment model (Baichuan-13B-Chat), has a strong dialogue ability, out of the box, a few lines of code can be easily deployed.

More efficient reasoning: in order to support the use of the majority of users, the project also opened up quantitative versions of int8 and int4, compared to the non-quantitative version in the case of almost no effect loss greatly reduced the deployment of machine resources threshold, can be deployed in consumer-grade graphics cards such as Nvidia RTX3090.

Open source, free and commercially available: Baichuan-13B is not only completely open to academic research, developers can also apply for free commercial use by email and obtain an official commercial license.

At present, the model has been published on HuggingFace, GitHub and Model Scope, and interested CTOnews.com friends can go to learn about it.

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.