v0.9.16

Release 0.9.16

New features

• Support setting shared memory for training job. #1104
• Support setting the init-container-image for pytorch-operator. #1097
• Add env-from-secret to read environment variables from secret. #1107
• Support config security context for KServe. #1108

Bug fixes

• Fix get kserve job panic. #1086
• Fix gpu-exporter and prometheus demo. #1087
• Fix incorrect TensorBoard images. #1109

Misc

• Update Arena Java SDK fastjson version. #1088
• Remove vendor. #1089
• Change kserve prom svc to ClusterIP. #1096
• Upgrade the kubernetes dependencies to v1.28 and go version to 1.21. #1102
• Add a demo for using arena CLI in container. #1105

Please follow the Get started Guide to install.

v0.9.15

Release 0.9.15

New features

• KServe support expose metrics automatically by --enable-prometheus & --metrics-port. #1073
• KServe support autoscaling according custom metrics by HPA. #1073

Bug fixes

• Fix port cannot be allocated when submitting a tfjob using the go sdk. #1071
• Fix --command parameter is not effective. #1074
• Fix command includes quotes cause Helm template failure. #1075

Misc

• Upgrade helm version to v3.13.3. #1072

Please follow the Get started Guide to install.

v0.9.14

Release 0.9.14

Arena now supports model management. You can use the arena model subcommand to manage registered model and model versions in mlflow, and associate them with your training jobs and serving jobs. For more information, please refer to Model Manage Guide.

New features

• Add support for MLflow model manage. #1058
• Add model manage documenation. #1066

Breaking changes

• Migrate model subcommand to model analyze. #1060

Misc

• Fix readthedocs build failed. #1069

Please follow the Get started Guide to install.

v0.9.13

Release 0.9.13

New features

• Add backend param for triton serving. #1039
• Support for updating the nodeSelector and toleration in GO SDK. #1043
• Support update --data in kserve serving job. #1049
• Support config request resources in kserve runtime. #1050

Bug fixes

• Delete cm if job failed. #1051

Misc

• Upgrade Kubernetes version 1.26.4 and go version 1.20.12. #1042

Please follow the Get started Guide to install.

v0.9.12

Release 0.9.12

New features

• Compatible with training-operator CRD. #1024
• Update tritonserver base image #1036

Bug fixes

• Fix KServe inferenceservice templete. #1034
• Fix the abnormal status of training jobs. #1011

Misc

• Add CI to run Go unit test. #1035
• Add CI to run the tests for Go. #1031

Please follow the Get started Guide to install.

v0.9.11

Release 0.9.11

Changed

• Update dependent component version.
• Support KServe inference service.
• Support maxSurge, livenessProbe, readinessProbe.

Please follow the Get started Guide to install.

v0.9.10

Release 0.9.10

Changed

• Fix --data-dir is not taking effect in custom-serving.
• Fix the prompt content when submitting serve job.
• Default delete secret permissions in et-operator.
• Enable create secret for deepspeedjob, etjob.

Please follow the Get started Guide to install.

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.