This is a Chinese instruction-tuning lora checkpoint based on llama-7B(1epoch) from this repo's work. Specially, this is the 4bit version trained with qlora