Pith. sign in

← back to paper

Review history

arxiv: 2506.13541 · 2 revisions

Mixture of Weight-shared Heterogeneous Group Attention Experts for Dynamic Token-wise KV Optimization

  1. 2026-08-15 CONDITIONAL HIGH v1.4.0-alltime-deepseek-medium novelty 6.0
    133896 ms 15233 in 14845 out 2026-08-15T19:58:57.418624+00:00
  2. 2026-08-07 CONDITIONAL MODERATE v1.4.0-alltime-deepseek-medium novelty 6.0
    132771 ms 15260 in 13960 out 2026-08-07T00:31:02.242045+00:00