[2410.10254] LoLCATs: On Low-Rank Linearizing of Large Language Models