Usage statistics or reports published by Meta or independent analysts

Meta Launches Llama 3 LLM: 8B and 70B Models, Trained on 15T+ Tokens

Apr 18, 2024, 04:37 PM

Meta Platforms has officially released Llama 3, a new version of its large language model (LLM), which includes 8B and 70B parameter models. These models are trained on over 15 trillion tokens with a knowledge cutoff in December 2023, significantly larger than their predecessor, Llama 2, and are designed to enhance language understanding and task performance across various applications. The release marks a significant step in the AI industry, positioning Meta as a leader in open-source LLM technology. The models are available for both text and image training, although the current versions are text-only. Meta also plans to integrate these models into its Meta AI virtual assistant and expand their functionality to include image outputs in the future, with a standalone website for the assistant.