Finetuned LLaMA-7B with LoRa using 82K Self-Instruct instances.