Llama 3 models trained on the tulu dataset, following https://arxiv.org/abs/2311.10702 (tulu 2) and https://arxiv.org/abs/2406.09279 (tulu 2.5).