This llama[1] model was trained 2x faster with Unsloth and Huggingface's TRL library.
[1] A. Grattafiori et al., The Llama 3 Herd of Models. 2024. [Online]. Available: https://arxiv.org/abs/2407.21783