Finetuned LLaMA-7B with LoRa using Self-Instruct instances. The instances were filtered using relevancy classifier targeting alignment between instruction + input and output.