Appendix E: Parameter-efficient Finetuning with LoRA