Meta Unveils New Llama 3 AI Models with Impressive Features

Meta Llama 3 AI Models Overview

Meta recently introduced its latest artificial intelligence models, Llama 3 8B and 70B, which are part of the Large Language Model Meta AI series. These new models boast enhanced capabilities compared to their predecessors and have been trained using innovative methods to improve their efficiency. Surprisingly, while the largest model in the previous generation was 70B, the new models will contain over 400 billion parameters. Meta plans to release smaller AI models in April and larger ones later in the summer.

Availability of Meta Llama 3

Meta is taking a community-first approach by making the Llama 3 models open source, just like its previous models. These new models will be available on various platforms such as AWS, Google Cloud, Microsoft Azure, and more. Meta has also integrated Llama 3 with its own Meta AI, accessible through Facebook Messenger, Instagram, and WhatsApp in supported countries.

Performance and Architecture of Meta Llama 3

In terms of performance, Meta shared benchmark scores for both the pre-trained and instruct models of Llama 3. The pre-trained model of Llama 3 70B outperformed Google's Gemini 1.0 Pro in various benchmarks, while the instruct model outscored the Gemini 1.5 Pro model in different tests. Meta has made improvements to the architecture of the new AI models, including using a tokeniser with a vocabulary of 128K tokens and implementing grouped query attention (GQA) to enhance inference efficiency. The models have been pre-trained with a vast amount of data sourced from publicly available sources.