Meta releases Llama 3 - 8B and 70B open LLMs

AI Impact Summary

Meta has released Llama 3, a new open large language model family, offering two sizes (8B and 70B) with base and instruction-tuned variants. The model’s key improvements include a larger vocabulary (128,256 tokens), Grouped-Query Attention (GQA) for the 8B model, and training on a significantly larger dataset (15 trillion tokens). This release provides developers with a powerful, openly accessible LLM for a variety of applications, particularly those leveraging the Hugging Face ecosystem.