BlueLM
About
BlueLM is a diverse family of large language models developed by vivo AI Lab, designed to be open-source and versatile. The models come in different sizes, including 7B parameter base and chat models, and are available with context lengths of both 2K and 32K. They were rigorously trained on a dataset comprising 2.6 trillion tokens, featuring a primary focus on Chinese and English languages, with additional data in Japanese and Korean. Notably, the BlueLM-7B-Chat model has demonstrated strong performance on benchmarks such as C-Eval and CMMLU. These models are licensed for use in both academic research and commercial sectors, making them highly flexible for a range of applications. Additionally, a 4-bit quantized version of the chat model is offered, which is optimized for reduced GPU memory usage 26.