New open-source model with 8k context runs on CPU, outperforms GPT-3

"MPT-30B...that is significantly more powerful than MPT-7B and outperforms the original GPT-3."

Source: 1. Original announcement: https://www.mosaicml.com/blog/mpt-30b