v0.9.9

Release 0.9.9

Changed

• Update SDK and JAVA SDK Unit test.
• Fix panic when pod started failed.
• Support job set image pull policy.
• Support new training type deepspeed.
• Fix evaluator node selector.
• Fix update serve duplicate create env and toleration.

Please follow the Get started Guide to install.

v0.9.8

Release 0.9.8

Changed

• Support Cron tfjob set ttlAfterFinished.
• Add DeepSpeed base image dockerfile.
• Move policy v1beta1 to v1.
• Fix evaluatejob job yaml in charts.

Please follow the Get started Guide to install.

v0.9.7

Release 0.9.7

Changed

• Support set TTLSecondsAfterFinished in Builder.

Please follow the Get started Guide to install.

v0.9.6

Release 0.9.6

Changed

• Add ownerReference for configmap and tensorboard.

Please follow the Get started Guide to install.

v0.9.5

Release 0.9.5

Changed

• Add imagePullSecret and shareMemory for arena serve.
• Support TTLSecondsAfterFinished.
• Support TFJob StartingDeadlineSeconds.
• Support TFJob/PytorchJob ActiveDeadlineSeconds.

Please follow the Get started Guide to install.

v0.9.4

Release 0.9.4

Changed

• Fix serve update when limits is null.
• Fix arena serve update bug.
• Fix model serve args bug.
• Add toleration dedup for arena serve update.

Please follow the Get started Guide to install.

v0.9.3

Release 0.9.3

Changed

• Optimize arena submit etjob for spot.
• Fix arena serve update bug.
• Fix model serve args bug.
• Add toleration dedup for arena serve update.

Please follow the Get started Guide to install.

v0.9.2

Release 0.9.2

Fixed

• Fix serve triton bugs.

Changed

• Skip to update crd when upgrade arena-artifacts.
• Modify the support method of Toleration.

Added

• Update images and support clean all policy for tfjob.
• Support submit parameters useHostNetwork useHostIPC useHostPID.
• Support for arena serve update.
• Support custom scheduler name.

Please follow the Get started Guide to install.

v0.9.1

Release 0.9.1

Fixed

• Fix the bug that failed to run pytorchjob with RDMA.
• Fix the bug that error dispaly gpu core resources on nodes.
• Fix the bug that add evaluator and tensorboard to pod group.

Changed

• Refact installtion.
• Modify restful-serving to http-serving of deployment services.
• Optimize the operators to omit the Completed jobs into the queue.

Added

• Support modeljob adapts helm3.
• Cron workload supports custom labels.
• Java SDK submits training job with --label.
• Add resource limits for tfjob.
• Add subpathexpr for job .

Please follow the Get started Guide to install.

v0.9.0

Release 0.9.0

Fixed

• Fix the bug arena update serving with specified kubeconfig.
• Fix the bug evaluatejob status not return.
• Fix the bug not set default shell type in arena client.
• Fix the bug of install arena when kubedl-operator existed.
• Fix the bug of mpi-operator crash.

Added

• Add command 'arena model' to support model profile/benchmark/optimize/evaluate before deploy.
• Mark 'arena evaluate' as deprecated as it has merged to 'arena model evaluate'.
• Upgrade git-sync image version to support git token.
• Upgrade arena java sdk to the latest version.
• Support execute shell with custom shell type like sh or bash.
• Support --clean-task-policy for mpijiob.
• Add arena-artifacts to adapt k8s 1.22.
• Support prometheus url token.
• Upgrade the helm version to v3.7.2 and kube client version to v1.23.0.

Please follow the Get started Guide to install.