Revolutionizing Natural Language Processing: Llama 3 Unveiled by Meta

Meta has recently unveiled its latest generation of open-source large language model, Llama 3, with two text-based models trained with 8B and 70B parameters. These models have been evaluated against other industry benchmarks and have shown state-of-the-art performance in reasoning, code generation, and instruction capabilities.

Llama 3 has undergone rigorous filtering processes to ensure data quality and reliability. It has been trained with a data set seven times larger than its predecessor, using more than 15T tokens collected from publicly available sources. Additionally, over 5% of the pre-training data set includes information in languages other than English, preparing for upcoming multilingual use cases.

Meta has introduced resources like Llama Guard 2, Code Shield, and CyberSecEval 2 to promote safe interactions with the models. The company plans to introduce new capabilities such as longer context windows, additional model sizes, and improved performance for Llama 3 in the coming months. Research work on the model will also be shared by the company.

To enhance user experience and capabilities across various platforms like Facebook, Instagram, WhatsApp, and Messenger, Llama 3 has been integrated into Meta AI assistant. Users can download these models on various platforms including Amazon Web Services, Google Cloud, IBM WatsonX, Microsoft Azure among others.