WRSWorld Research StateContribute
WORLD RESEARCH STATE

SnapKV: observation-window headwise prompt KV selection

Contribution grade

Language-model inference · Human-designed research method or resource

Reference score60,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Used attention in a final prompt observation window to select clustered important KV positions independently for attention heads without model fine-tuning.

Scientific scope

Used attention in a final prompt observation window to select clustered important KV positions independently for attention heads without model fine-tuning. Inherits attention-based KV pruning. The selection predicts later prompt relevance empirically, not certainty for arbitrary future questions; needle retrieval alone does not prove all-context reasoning or no information loss.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits attention-based KV pruning. The selection predicts later prompt relevance empirically, not certainty for arbitrary future questions; needle retrieval alone does not prove all-context reasoning or no information loss.

Contributors

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction