This model is intended to be a strong base suitable for downstream fine-tuning on a variety of tasks. Based on our internal evaluations, we believe it's one of the strongest models for most down-stream tasks. You can read more about our development and evaluation process
here.
It is a hierarchichal SLERP merge of teknium/OpenHermes-2.5-Mistral-7B, Intel/neural-chat-7b-v3-3, meta-math/MetaMath-Mistral-7B, and openchat/openchat-3.5-1210. berkeley-nest/Starling-LM-7B-alpha was omitted from this version of the model.
