"MPT-7B looks to be super competitive across the board, even beats 13B models. This LLM is trained on 1T tokens of text and code curated by MosaicML. The model is fine-tuned to also work with a context length of 65k tokens!"