Llama 3, the latest generation of Meta’s open-source large language model, has been unveiled. The models have been trained with 8B and 70B parameters, making them best-in-class open source models for a wide range of use cases. These models offer improved reasoning capabilities and are integrated into the Meta AI assistant.
Meta aims to continue driving innovation in AI across various applications, tools, and optimizations. Two text-based models trained with 8B and 70B parameters have been introduced, showcasing state-of-the-art performance and improvements in reasoning, code generation, and instruction capabilities compared to their predecessors. These models have undergone rigorous filtering processes to ensure data quality and reliability.
The pre-training data set for Llama 3 is seven times larger than its predecessor, using more than 15T tokens collected from publicly available sources. More than 5% of the pre-training data set includes information in languages other than English, preparing for upcoming multilingual use cases. To ensure responsible use of Llama 3, Meta has introduced resources like Llama Guard 2, Code Shield, and CyberSecEval 2 to promote safe interactions with the models.
Llama 3 has been integrated into the Meta AI assistant across platforms like Facebook, Instagram, WhatsApp, and Messenger. Users can download Llama 3 models soon on various platforms including Amazon Web Services, Google Cloud, IBM WatsonX, Microsoft Azure