← back to paper

Review history

arxiv: 2605.15422 · 2 revisions

DualKV: Shared-Prompt Flash Attention for Efficient RL Training with Large Rollouts and Long Contexts

1. 2026-06-30 UNVERDICTED LOW v0.9.1-grok novelty 6.0
   25107 ms 5949 in 1412 out 2026-06-30T20:56:18.971338+00:00
2. 2026-05-19 CONDITIONAL LOW v0.9.0 novelty 8.0
   51921 ms 5949 in 1326 out 2026-05-19T15:44:11.884822+00:00