Low-Rank Approximation for Sparse Attention in Multi-Modal LLMs | Read Paper on Bytez