Quan Chen

Native sparse attention has emerged as a promising approach for efficient long-context LLM inference without compromising accuracy. While it significantly reduces the attention computation and KV cache access costs, the KV cache size exhibits steeper linear growth with the context length. As a re...

Cold memory rebalancing exhibits unique challenges to the existing memory overcommitment mechanisms, such as page tracking under Transparent Huge Pages (THP) and frequent remapping of page tables, leading to a series of throughput and latency issues. In this paper, we propose Blowfish, a memory o...