Llama-3-8B-Instruct is Meta's compact dialogue model with 8B parameters, trained on 15T tokens. This instruction-tuned decoder demonstrates competitive performance against larger closed-source models in human evaluations.