Lora weight for Llama2 trained on oasst1 dataset using 4bit quantazation QLoRA.