Post-training of Vision-Language-Action (VLA) models is crucial due to the diversity of simulators, robot embodiments, and task objectives. Existing compute services, whether direct accelerator rentals or batch workload submissions, typically allocate exclusive GPU and CPU resources to a single tenant. While this maximizes client flexibility, it burdens users with infrastructure adaptation, and the fixed card-hour accounting model makes short or bursty workloads expensive for tenants and inefficient for service providers.
To address these challenges, we present JoyNexus, a unified service for multi-tenant VLA supervised fine-tuning, reinforcement learning, and evaluation. JoyNexus decouples the Training Model Service, Inference Model Service, and Environment Service, each accessed through APIs and backed by resident shared base models with tenant-specific slots. Tenants can invoke high-level semantic APIs for training, rollout, and evaluation, or compose custom algorithms using lower-level APIs and their assigned endpoints. Multiple tenants can submit workloads concurrently; their action modules, optimizers, rollout records, and policy versions remain isolated, scheduled by the global Training Queue and Inference Queue.
To further improve multi-tenant training efficiency, JoyNexus introduces group batching for heterogeneous VLA data schemas that share a compatible model-facing prefix, enabling a single shared backbone forward pass over grouped samples.
Finally, we evaluate JoyNexus through workload simulation and a group-batching pipeline in a realistic embodied scenario. Results show that, compared with isolated single-tenant execution, JoyNexus reduces aggregate GPU time and improves service utilization via cross-tenant scheduling on shared resources.
Blogger's Review: JoyNexus ingeniously addresses resource allocation and utilization efficiency issues in multi-tenant environments. By decoupling services and introducing group batching, it reduces training costs and enhances service performance. This innovation promises strong support for future VLA model training with broad application prospects.