Meta's open-source language model with fully released training code and logbook — complete transparency.
Open LLaMA-recipe training data — 30T tokens, fully transparent
