Insights from Finetuning LLMs with Low-Rank Adaptation