Oriel — AI & technology worth noticing, every day.

Support New AI Frameworks Faster Using NVIDIA Workload Map

Web page · https://developer.nvidia.com/blog/support-new-ai-frameworks-faster-using-nvidia-workload-map

NVIDIA DSX AI Factory Platform brings together technologies to help the ecosystem design, build, and operate more efficient AI factories, including open source,...

First seen by Oriel on .

AI summary

NVIDIA Workload Map uses declarative Kubernetes definitions to make distributed workloads reusable across inspection, gang scheduling, platform management, and multi-cluster scheduling without framework-specific adapters.

  • Workload Map ships as a Kubernetes CRD and Go library that describe pod templates, replicas, hierarchy, metadata, and status for each workload type.
  • A JobSet definition identifies leader and worker roles, derives status with jq expressions, and calculates pod counts from replicas multiplied by parallelism.
  • The example CLI resolves one JobSet into three pods and aggregates its resource request to 12 GPUs, including separate leader and worker components.
  • KAI Scheduler uses definitions for gang-scheduling fallback, while NVIDIA Run:ai uses the same catalog entry to manage the JobSet as one workload.
  • The Apache 2.0 project includes definitions for JobSet, PyTorchJob, Ray, KServe, Knative, MPIJob, NVIDIA NIM, LeaderWorkerSet, Milvus, Grove PodCliqueSet, and NVIDIA Dynamo.

#kubernetes #workload scheduling #ai factories #distributed training

Generated by gpt-5.6-sol. Check the original for details.

Aliases

SourceExternal IDURL
NVIDIA Developer Bloghttps://developer.nvidia.com/blog/?p=124270https://developer.nvidia.com/blog/support-new-ai-frameworks-faster-using-nvidia-workload-map/ Read discussion