Skip to content

Release 0.4.0

Added

  • Add Pytorch
  • Support custom serving
  • Add tarball installation for Linux and Mac
  • Support non-root installation
  • Add train init framework
  • Add GPU support for PS
  • Support GangScheduling Native in MPIJob
  • Support GangScheduling Native in PytorchJob

Fixed

  • Upgrade Deployment version from extensions/v1beta1 to apps/v1
  • Fix the issue of incorrect number of allocated GPUs
  • Upgrade Helm to v2.14.1
  • Fix evaluator & chief validation
  • Fix incorrect cpu resource variable, should be psCPU
  • Set exit code as 2 when delete job failed
  • Fix the bug of using Estimator
  • Fix the bug of deploying Prometheus
  • Support Kubernetes 1.18 and above