Who Should Score the Cache? Mid-Depth Layers Are the Best Judges of KV-Cache Eviction
Ekaterina Seregina
実装難易度
Hard
推論・学習コスト
Low
想定用途
QA
概要
Key-value (KV) cache eviction methods rank tokens by attention scores, and each layer scores its own cache. To our knowledge, no eviction method asks which layer should score. We find, across twenty-one models, that the depth of the scoring layer governs eviction quality. On the eighteen models with per-layer caches, spanning 0.5B to 109B, long-context retrieval accuracy collapses at shallow…
何が新しいか
Key-value (KV) cache eviction methods rank tokens by attention scores, and each layer scores its own cache. To our knowledge, no eviction method asks which layer should score.
何に使えるか
QA
実装情報
- Paper URL
- あり
実装チェックリスト
実装または配布ページ
要確認Paper onlyの可能性があるため再実装前提で確認してください。
一次情報リンク
OKPaper
検証しやすさ
要確認公式実装が見つからないため、論文から再実装する前提です。
計算資源
OK小規模データならCPUまたは単一GPUで検証しやすい領域です。
ライセンス
未取得配布元のLICENSE、モデルカード、Paperの利用条件を確認してください。
商用利用
未取得研究利用限定、データセット由来制限、API規約の有無を確認してください。
自社データで試すなら
製造業・材料開発のExcel/CSVデータに落とし込むための最初の手順です。
- 1まず自社データを、入力条件、目的変数、評価したい指標に分けて整理します。
- 2LightGBMやRandom Forestなどのベースラインを先に作り、この手法と比較します。
- 3評価指標はR2/RMSE、AUC、異常検知の再現率、実験回数削減率など、現場の意思決定に近いものを選びます。
- 4SHAPや特徴量重要度で、効いている因子が物理・化学・工程知識と矛盾しないか確認します。
実装難易度
Hard - 公式実装が見つからないため、論文から再実装する前提です。
必要リソース
- GPU目安: Low
- データセット: 論文・リポジトリ側の指定を確認してください。
- 学習要否: 再学習や評価環境の準備が必要になる可能性があります。
- 小規模データならCPUまたは単一GPUで検証しやすい領域です。
実務で使う場合の注意点
- ライセンスと商用利用条件は、Paper / GitHub / Hugging Face の配布元で確認してください。
- 精度、再現性、計算コストはデータセットや評価条件に依存します。
- 個人情報や機密データを扱う場合は、入力データの保存先と外部API利用条件を確認してください。
関連記事
NitiBench+: A Temporal Thai Legal Question Answering Benchmark Across 17 Domains
Law changes over time. A court decides a case under the law in force on that day, and legal QA benchmarks grad
CREDO: Evidence-Conditioned Credit Assignment for Temporal Reasoning Agents
Temporal Knowledge Graph Question Answering (TKGQA) requires language model agents to retrieve evidence, selec
R^2Orch: Recursive Self-Improvement for Agent Orchestration via Self-Play within a Skill-Internalization Loop
In recent years, agentic orchestration has been widely applied to solve complex human tasks. However, existing
PACT: Certifying What Is Written in Parallel Decoding of Diffusion Language Models
Parallel decoding speeds up masked diffusion language models by writing several tokens per forward pass. Publi