This repository is designed to make it easy to train and evaluate llama-2 models. The default settings can be ran for free on Google Colab. Follow the instructions in the QLoRA_Llama2 notebook to train your own model. Follow the instructions in the Evaluation_harness notebook to evaluate your model. By default we evaluate on the MMLU dataset. The model that resulted from my own training run are available here: https://huggingface.co/vincentmin/llama-7b-orca/tree/main