#machine learning
May 2026
EntmaxKV: Support-Aware Decoding for Entmax Attention
EntmaxKV is introduced, an entmax-native sparse decoding framework that exploits sparsity before KV pages are loaded, and a Gaussian-aware entmax selector that estimates the entmax threshold from lightweight page statistics, adapting the selected budget to the score distribution.
Gonçalo Duarte, Miguel Couceiro, Marcos V. Treviso
· arXiv.org · 0 citations