SPIN Unprocessed September 4, 2026 ai_technology research
What Matters for Aggressive Decoding-Time KV Eviction? Temporal Aggregation and Ranking Preservation
View original on arxiv.orgOverview
arXiv:2609.03515v1 Announce Type: new Abstract: Decoding-time KV cache compression research focuses heavily on designing better token scoring functions, while the temporal rule that aggregates scores across decode steps is often treated as an implementation detail. Under aggressive KV compression, we find that exponential-moving-average (EMA) aggregation makes approximately order-preserving scorer modifications largely indistinguishable at the eviction-set level. Value-norm and entropy variants
SpinGraph analysis pending — check back after processing.
Ask AI about this story
Opens with the SpinGraph .md URL and structured context — one click, prompt included.
More from arXiv Artificial Intelligence
View all →- GPS-Bench: A Governance Policy Benchmark for Automating Policy Analysis
- Dalek: A Constructive Agent Machine
- Feature Reconfiguration With Visual Prior for Medical Lesion Segmentation
- NeoRed: A Knowledge-Logic-Alignment Multimodal Large Language Model for Neonatal Respiratory Disease Diagnosis
- CulturalMenuBench: Probing the Knowledge-Application Gap in Multimodal Culinary Reasoning
- PPO-STGNN: A Proximal Policy Optimization Approach with Spatio-Temporal Graph Neural Networks for DAG Task Scheduling in Cloud-Edge-End Computing
Markdown (.md) · JSON-LD schema (.json) · Machine-readable for AI & GEO