Meta releases Llama 3.1 - 405B, 70B & 8B with long context

AI Impact Summary

Meta has released Llama 3.1, a new family of large language models with significant improvements including a 128K token context length, multilinguality, and tool usage capabilities. The models are available in 8B, 70B, and 405B parameter sizes, offering flexibility for various deployment scenarios. Notably, the 405B model enables synthetic data generation and distillation, expanding use cases for this model. The release includes detailed memory requirements for both training and inference, highlighting the need for substantial GPU resources, particularly for the larger models.