Llama 3: Meta's Latest LLM Breakthrough!
Llama 3 has just hit the scene, marking a significant advancement in open Large Language Models (LLM) by Meta. With a notable ~10% improvement over its predecessor, Llama 3 arrives in two sizes, 8B and 70B, featuring an extended tokenizer and a commercially permissive license, opening doors to diverse applications. Explore the groundbreaking features of Llama 3: Trained on 15 trillion tokens and fine-tuned on 10 million human-annotated samples. Available in 8B and 70B versions as Instruct and Base models. Llama 3 70B emerges as the top-performing open LLM on the Multimodal Language Understanding (MMLU) task. Instruct model excels in coding tasks with Human Evaluation scores of 62.2 for 8B and 81.7 for 70B. Introducing a Tiktoken-based tokenizer with a 128k vocabulary. Default context window set at 8192, with the option for expansion. Utilizes SFT, PPO, and DPO for alignment. Commercial use permitted, making it accessible for various applications. Available on Hugging Face with 1-click deployments on platforms like Amazon SageMaker and Google Cloud. Stay tuned for upcoming model sizes and enhanced performance, as Llama 3 continues to push the boundaries of AI innovation.