Built on a Mistral base again this time, this is a SLERP merge of Envoid/Lumiere-7B and NousResearch/Capybara-7B-V1.9-Preview. Looks like this model produces pretty decent prose, and has good coherency and story awareness, above and beyond base Mistral models.

Capybara on it's own is pretty decent here, but I think the little bit of scientific data in the former provices a touch extra prose punch. This feels usable in the way 13b models are to me (v2 does too actually if you prefer a llama-2 model)