Today’s release includes 8B, 70B and 405B Llama 3.1 models that were trained on a greater quality and quantity of data than Llama 3 for both pre- and post-training. All three models were trained on over 15T tokens.

→ View original post on X — @aiatmeta