Baichuan
About
Baichuan 2, developed by Baichuan Intelligent Technology, is an advanced series of open-source large language models. It has been trained on a substantial corpus of 2.6 trillion tokens, showing remarkable performance across various Chinese and English benchmarks. The model family comprises 7B and 13B parameter versions, offered in both base and chat configurations, and includes a 4-bit quantized chat model for efficient deployment. These models are available for both academic research and commercial use under an official license. Leveraging PyTorch 2.0's F.scaled_dot_product_attention, they ensure faster inference. Additionally, intermediate checkpoints from its training are provided, supporting research into model training dynamics 23.