AI RESEARCH
DualKV: Shared-Prompt Flash Attention for Efficient RL Training with Large Rollouts and Long Contexts
arXiv CS.LG
•
ArXi:2605.15422v1 Announce Type: new Modern RL post-