.png)
Meta has unveiled the third generation of its Llama AI model, boasting impressive advancements. The current versions feature 8 billion and 70 billion parameters, with plans for an even larger model with 400 billion parameters in the works. Currently in the training phase, these models are poised to revolutionize AI capabilities.
According to Meta, Llama 3's training dataset is seven times larger than its predecessor, Llama 2, and contains four times as much code. Moreover, the tokenizer used for encoding language has been significantly improved for enhanced performance. Meta has also implemented a technique called 'grouped query attention' to optimize efficiency.
However, despite these advancements, 95 percent of the training data is in English, resulting in diminished performance for the thirty other languages supported by Meta AI. Nevertheless, Llama 3 will be integrated into services such as WhatsApp and Instagram as an AI assistant. Currently available only in English, this feature is accessible in select countries including the United States, Australia, Canada, Ghana, Jamaica, Malawi, New Zealand, Nigeria, Pakistan, Singapore, South Africa, Uganda, Zambia, and Zimbabwe.
English (US)