FlashOffloading: Full KV Cache in Host DRAM for Long-Context Serving

Xuan Wang, Zhuanhao Wu, … Cheng Li (corresponding) · Huawei & Shanghai Jiao Tong University · 2026 · Ascend 910C, GLM-5.2
arXiv 2611.07542 Interactive page

References

  1. FlashOffloading: Low-Latency KV Cache Offloading for Long-Context LLM Serving (Wang et al., 2026)
  2. FlexGen (Sheng et al., 2023)
  3. FlashAttention-3 (Shah et al., 2024)
  4. DeepSeek-V3.2-Exp (DeepSeek-AI, 2025)
  5. Quest (Tang et al., 2024)
  6. Mooncake (Qin et al., 2024)
  7. ZeRO-Infinity (Rajbhandari et al., 2021)
  8. Native Sparse Attention (Yuan et al., 2025)
  9. InfiniGen (Lee et al., 2024)
  10. LMCache (Liu et al., 2025)
  11. Mooncake: Trading More Storage for Less Computation (Qin et al., 2025)
  12. ECHO (Liu et al., 2026)
  13. HiSparse (Xie et al., 2026)
  14. Understanding Bottlenecks for KV Offloading (Meng et al., 2025)