Meta’s Llama 3.3 multilingual LLM is an instruction tuned generative model in 70B that for some applications approaches the performance of Llama 3.1 405B and other frontier-level closed source models