Llama3 trained on our RP datasets
2.5 Not good
Pick which version of the model you base your review on.