PagedEviction prunes the KV cache in whole blocks using a key-value norm ratio, speeding up long-context LLM inference in vLLM while staying close to full-cache accuracy.
Proceedings of Machine Learning and Systems, 6: 196–209
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
PagedEviction: Structured Block-wise KV Cache Pruning for Efficient Large Language Model Inference
PagedEviction prunes the KV cache in whole blocks using a key-value norm ratio, speeding up long-context LLM inference in vLLM while staying close to full-cache accuracy.