AI RESEARCH
SparDA: Sparse Decoupled Attention for Efficient Long-Context LLM Inference
arXiv CS.LG
•
ArXi:2606.04511v1 Announce Type: cross Sparse attention reduces compute and memory bandwidth for long-context LLM inference. However, two key challenges remain: (1) KV cache capacity still grows with sequence length, and offloading to CPU memory