做强化学习持续训练的团队终于有了高效的并行方案——2.81 倍吞吐量提升且不损失奖励，直接开源可用，建议试试。
Trajectory 与 UC Berkeley Sky Lab 和 Anyscale 合作，构建了一个用于持续学习的并发多 LoRA 训练栈。该方案将每个强化学习实验映射到始终热运行的引擎上的专用 LoRA 适配器，相比单租户基线实现了 2.81 倍的端到端实验吞吐量提升，且无奖励回归。代码已在 NovaSky-AI/SkyRL 开源。这一进展解决了持续学习中多实验并行效率低下的问题，对强化学习研究和工程团队有直接价值。
Trajectory Releases a Concurrent Multi-LoRA Training Stack for Continual Learning, Reporting a 2.81× Experiment-Throughput Gain
Trajectory, working with UC Berkeley Sky Lab and Anyscale, built a concurrent multi-LoRA training stack for continual learning. It maps each RL experiment to a dedicated LoRA adapter on an always-hot engine, reporting a 2.81× end-to-end experiment-throughput gain over a single-tenant baseline with no reward regression. The code is open-sourced in NovaSky-AI/SkyRL. The post Trajectory Releases a Concurrent Multi-LoRA Training Stack for Continual Learning, Reporting a 2.81× Experiment-Throughput Gain appeared first on MarkTechPost .