MEGA Hub

SemPIC: Learning Semantic Position-Independent KV Caches

Authors

Do you know Hui Xie?You can claim authorship or link another user.Do you know Peng Xiao?You can claim authorship or link another user.Do you know Yutong Deng\textsuperscript?You can claim authorship or link another user.Do you know Shuoran Dou?You can claim authorship or link another user.Do you know Jian Yang?You can claim authorship or link another user.Do you know Jinyang Guo?You can claim authorship or link another user.

Abstract

Long-context retrieval and agentic workloads repeatedly reuse the same documents under changing instructions, histories, and document orders. Prefix caching cannot exploit this reuse, while position-independent caching (PIC) remains unreliable because independently compiled KV states lack the future context in which they will be consumed. Our diagnostics show that a learned boundary-conditioned baseline sharply reduces attention deviation near reusable-block boundaries but leaves interior and task-level residuals, motivating adaptation of the document representation itself. We present \emph{SemPIC}, which trains a LoRA-enabled Writer to compile native per-layer document KVs through behavioral distillation while retaining the pretrained decoder as an unchanged Reader. Adaptation is confined to offline cache construction, preserving the standard KV interface and cache-hit decoding path. We further introduce KV Gradient Checkpointing, which reduces peak training memory without severing gradients through cached KVs. Across three models and four tasks, SemPIC raises mean micro-F1 over KV Packet from 0.53 to 0.60, approaching Full Recompute at 0.62.

Community

00