Llama2-7b finetuned on a 10k subset of OpenOrca focusing on multiple choice questions + 6k ShareGPT4 datasets.