As a base model we used: https://huggingface.co/chavinlo/alpaca-13b
Finetuned on GPT4's responses, for 3 epochs.
NO LORA
