Urgent.News

What's breaking now, across thousands of outlets.

More in AI

Fault tolerant distributed training on Amazon EKS using NVRx

Integrate NVIDIA Resiliency Extension (NVRx) into PyTorch FSDP training on Amazon EKS to overlap checkpoint I/O with training and recover from GPU faults in seconds.

  • NVRx integrates into PyTorch FSDP training on Amazon EKS to improve fault tolerance.
  • Async checkpointing, in-process restart, and ftlauncher are NVRx's key features.
  • Tested on EKS cluster with H100 GPUs, achieving robust and efficient distributed training.

More from Wednesday 16 September →