Pith. sign in

By the same argument as in the proof of Lemma A.7, there exist functions ϕ:R d0 →R d1 andρ:R d1 →R dout such that for everya= (a 1

(Noncompressibility) For every KV cache compression policy C with compression budget satisfyingr(n)< n there exists a suffixb∈A k withk+n≤Nsuch that, ∥M([a,b])−M C,a(b)∥> C · arXiv 7800.4610

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it
read on arXiv browse 1 citing papers

citation-role summary

background 1

citation-polarity summary

background 1

fields

cs.LG 1

years

2026 1

verdicts

UNVERDICTED 1

roles

background 1

polarities

background 1

representative citing papers

Training Transformers for KV Cache Compressibility

cs.LG · 2026-05-07 · unverdicted · novelty 6.0 · 2 refs

Training transformers with KV sparsification during continued pretraining produces representations that admit better post-hoc KV cache compression, improving quality under memory budgets for long-context tasks.

citing papers explorer

Showing 1 of 1 citing paper.

  • Training Transformers for KV Cache Compressibility cs.LG · 2026-05-07 · unverdicted · none · ref 66 · 2 links

    Training transformers with KV sparsification during continued pretraining produces representations that admit better post-hoc KV cache compression, improving quality under memory budgets for long-context tasks.