JobSet

Headlines · 2

1. Nov 2
   NVIDIA open-sources automated recovery for distributed AI training

   NVIDIA released NVIDIA Recovery, an open-source workload-level service within NVIDIA DSX OS that automates recovery across distributed AI training jobs. It detects interruptions through application logs, GPU metrics, and scheduler signals, then attributes failures, isolates affected resources, and coordinates checkpoint restarts on healthy capacity. The service supports Slurm and can be deployed on an operator’s own cluster, while Kubernetes support remains on NVIDIA’s roadmap.

   NVIDIA Technical Blog · 🔥 0
2. Nov 2
   NVIDIA open-sources Workload Map for Kubernetes workloads

   In a technical blog post, NVIDIA introduced Workload Map, an open-source component licensed under Apache 2.0 that provides a reusable declarative description of Kubernetes workload structures across frameworks. A JobSet definition can support workload inspection, gang scheduling, status interpretation, submission, mutation and platform visibility. NVIDIA says KAI Scheduler uses Workload Map as a fallback for workload types without native plugins, NVIDIA Run:ai uses it in production, and Elotl Nova uses it to interpret workload structures and resource requirements during multi-cluster scheduling.

   NVIDIA Technical Blog · 🔥 0

Experience and discussion from the community

Share my JobSet experienceAsk about JobSet