Batched Low-Rank Adaptation of Foundation Models