Vivo self-developed Blue Heart Model 7B is open source to the whole industry, 2.6 trillion token (with download address)

CTOnews.com, November 9, at the recent 2023 vivo developer Conference, vivo announced that it was the first to open source 7 billion parameter-level Blue Heart Model 7B and its corresponding fine-tuning framework.

According to the promise, vivo is now officially open source on Hugging Face. BlueLM-7B,CTOnews.com partners can be downloaded from the vivo AI Lab home page.

BlueLM-7B open source large model includes 7B basic model and 7B dialogue model, while vivo open source supports 32K long text basic model and dialogue model.

According to reports, BlueLM uses a high-quality corpus for training, with a scale of 2.6 trillion token, which contains Chinese, English and a small amount of data from Japan and South Korea. Among them, BlueLM-7B-Chat achieved leading results in both C-Eval and CMMLU.

Both BlueLM-7B-Base-32K and BlueLM-7B-Chat-32K support 32K long text, which can support longer context understanding while maintaining the same basic capabilities.