Llama 2 is trained on 40% more public data and can process twice as much context than Llama 1, according to Meta.
