CUHK Team Open Sources Libra: 3x Throughput for Agentic Training

jiqizhixin · x · 2026-08-21

CUHK team released Libra to fix resource contention in agentic RL post-training. It uses a global planner to dynamically shift GPUs between rollout and training via an elastic hybrid pool. It also uses a causality-driven scheduler based on tool-return signals. Tested on 48 A800 GPUs, Libra achieves 3.0x higher throughput and converges 2.5x faster in reward.

Original post →

More from Infra

Infra channel →