Algebraically Identity-Preserving KV Tiering for Memory-Bounded Long-Context Inference
Published in Targeting EMNLP 2026 (ARR), 2026
Manuscript and artifact in progress. Project artifact: HALO.
Recommended citation: Ziqing Li, "Algebraically Identity-Preserving KV Tiering for Memory-Bounded Long-Context Inference." Targeting EMNLP 2026 (ARR).