Baichuan 2 13B

About

Baichuan 2 13B is a multilingual language model by Baichuan Intelligence Inc., built on a transformer decoder architecture. It features SwiGLU activations, memory-efficient xFormers attention, and Layer Normalization. Trained on 2.6 trillion tokens, it excels in various NLP tasks, such as text generation and translation, with strong performance in Chinese and English benchmarks. This 13B parameter model employs Attention with Linear Biases for enhanced efficiency. While open-source, commercial use requires a license, and it's evaluated as state-of-the-art on benchmarks like C-Eval and MMLU. Caution is advised against usage posing national security risks 24711.

Capabilities

Providers(1)