The Flow has not summarised this story yet — read it at arXiv Computation and Language.
The paper investigates when attention maps can be compressed beyond simple sparsity, arguing that large weights alone do not guarantee compressibility. It introduces metrics such as global score gaps and weighted sums of omitted values to determine token retention for a target mass, and presents a retrieval–aggregation model to predict the impact of truncation. Based on these insights, the authors propose CertKV, a training‑free compressor that allocates a tail‑summary slot per head and distributes remaining slots according to value dispersion, achieving strong performance across several benchmarks.
The paper addresses the challenge of long input contexts in Retrieval-Augmented Generation (RAG) systems, where concatenating many retrieved chunks increases prefill workload and time to first token (TTFT). It proposes a dual strategy: fine‑tuning the model to be aware of KV cache concatenation and selectively recomputing only part of the KV caches. Experiments on the RULER benchmark show that for a 124k‑token input, this combined method boosts the RULER score by 9.7 points over a baseline that recomputes caches only, while cutting TTFT by 80% compared with full attention.
NestedKV is a training‑free key‑only KV cache compression technique for long‑context language models that uses global, block‑level, and sliding‑window key anchors to score tokens via multi‑time‑scale cosine anomaly. It combines these rankings with a head‑adaptive outer learner and surprise‑gated token routing, requiring no model modification or additional training. Experiments on Qwen3 and Llama‑3.2 across benchmarks such as RULER, LongBench, and MMLU‑Pro show that NestedKV outperforms existing methods when the retained cache is small, achieving up to 19‑point gains on RULER and LongBench at a 75% retention rate.
arXiv:2608. 01247v1 Announce Type: cross Abstract: Query-agnostic KV cache eviction compresses a context once and reuses the resulting cache for arbitrary future queries, but performance can collapse under tight budgets.
Query-agnostic KV cache eviction compresses a context once and reuses the resulting cache for arbitrary future queries, but performance can collapse under tight budgets. Existing methods primarily improve which original KV pairs are retained.
arXiv:2607. 06519v1 Announce Type: new Abstract: Long-context LLM inference is increasingly limited by the memory and bandwidth cost of KV caches, yet aggressive compression can remove the layer-specific evidence needed for retrieval and multi-step reasoning.