github NVIDIA/KAI-Scheduler v0.5.1

latest releases: v0.10.0-rc3, v0.10.0-rc2, v0.10.0-rc1...
5 months ago

What's Changed

  • Scheduling gates by @itsomri in #122
  • [Bugfix] Reclaim victim queue order by @itsomri in #62
  • make nodeSelector, affinity and tolerations configurable by @gflatters in #127
  • Replace run.ai string with kai.scheduler by @omer-dayan in #131
  • Made scheduling queue label key configurable by @romanbaron in #129
  • Cache GetDeservedShare and GetFairShare by @davidLif in #139
  • refactor(binder): add cache to the resource reservation client by @enoodle in #144
  • Remove requirement for Worker when using PyTorchJob by @Phlip79 in #149
  • PodGroup info caching for some of the results by @davidLif in #138
  • refactor(scheduler): patch pod labels concurrently by @enoodle in #147
  • [Design] Minimum runtime before preemptions and reclaims by @ArmedGuy in #126
  • scheduler: bugfix: Make pod_scenario_builder build scenarios for rest of elastic job by @ArmedGuy in #132

New Contributors

Full Changelog: v0.5.0...v0.5.1

Don't miss a new KAI-Scheduler release

NewReleases is sending notifications on new releases.