JigsawRL: Assembling RL Pipelines for Efficient LLM Post-Training
For practitioners deploying RL-based LLM post-training, JigsawRL improves resource utilization and throughput with moderate latency trade-off.
JigsawRL introduces Pipeline Multiplexing for RL post-training of LLMs, achieving up to 1.85x throughput over Verl on synchronous RL and 1.54x over StreamRL/AReaL on asynchronous RL across 4-64 GPUs.
We present JigsawRL, a cost-efficient framework that explores Pipeline Multiplexing as a new dimension of RL parallelism. JigsawRL decomposes each pipeline into a Sub-Stage Graph that exposes the intra-stage and inter-worker imbalance hidden by stage-level systems. On this abstraction, JigsawRL resolves multiplexing interference through dynamic resource allocation, eliminates fragmented utilization by migrating long-tail rollouts across workers, and formulates their coordination as a graph scheduling problem solved with a look-ahead heuristic. On 4-64 H100/A100 GPUs across different agentic RL pipelines and models, JigsawRL achieves up to 1.85x throughput over Verl on synchronous RL, 1.54x over StreamRL and AReaL on asynchronous RL, and supports heterogeneous pipelines with moderate latency trade-off.