← back to paper
arxiv: 2607.01299 · 2 revisions
HYPIC: Accelerating Hybrid-Attention LLM Serving with Position-Independent Caching