Meta Llama 3.3 is a state-of-the-art 70 billion parameter multilingual large language model (LLM) designed for text generation tasks. As an instruction-tuned variant of the Llama architecture, it specializes in assistant-like dialogue applications across English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai. The model employs an optimized transformer architecture with Grouped-Query Attention (GQA) for efficient inference, trained on over 15 trillion tokens of publicly available data with a knowledge cutoff in December 2023. It leverages both supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align responses with human preferences for helpfulness and safety. Notable features include a 128k token context window, tool calling capabilities, and compliance with Meta’s custom commercial license (Llama 3.3 Community License). The model demonstrates strong performance on industry benchmarks while explicitly prohibiting unlawful uses or applications in unsupported languages without proper safety measures.
