Pretty sure openchat-3.5 is a mistral fine tune as well. The trick is not this neural alignment - it is training on many, many more tokens than Chinchilla reco

Pretty sure openchat-3.5 is a mistral fine tune as well. The trick is not this neural alignment - it is training on many, many more tokens than Chinchilla recommends.