RiseML v1.2.3 (2018/05/09)
This release changes behaviour for failing hyperparameter experiments and pins the minio version.
- Hyperparameter experiments no longer fail when a (sub-) experiment fails. You can also kill single (sub-) experiments.
- The minio version is now pinned to RELEASE.2018-04-19T22-54-58Z, it introduced a regression bug in the latest versions
RiseML v1.2.2 (2018/04/09)
This release adds support for private registries on GKE and improves scheduling capabilities.
- support private registries on GKE, automatically detected, no configuration needed
tolerations.trainingto better control where training and build jobs can be run
- fix for TensorBoard redirects being wrong in some cases
- fix for Postgresql deployment failing when hidden directories in mount path present
- The default value for
postgresql-db. Please make sure to use the value you used during your installation, i.e. add
postgresql.persistence.subPath: ""to your configuration if it was previously not set.
RiseML v1.2.1 (2018/03/27)
This fixes a bug when upgrading from a previous version.
- fix for Kubernetes sync service which was still using nodePorts which are no longer needed
RiseML v1.2.0 (2018/03/22)
riseml rmfor managing your data and the output generated by your experiments. See how they work in our docs.
nodeSelectors.riseml: A new Helm config variable to tell RiseML to only install and run on Kubernetes nodes with a specific label!
- Improved support for nvidia-docker2 and Nvidia's device plugin
- Fix for gpu pools on GKE that added a node taint automatically
- Fix for trailing '' in your run commands
RiseML v1.1.0 (2018/02/21)
This release brings Horovod support, a framework for distributed training with TensorFlow, as well as support for NVidia's K8s device plugin and GKE.
- Horovod support
- Support for Nvidia's device plugin for Kubernetes - just install it and configure your RiseML installation to use
- Fixes when building images on GKE
- Several UTF-8 fixes when building images and printing training logs
RiseML v1.0.4 (2018/02/01)
This release fixes Tensorboard when using Tensorflow >= 1.4.0. NOTE: At the same time, this means, that Tensorboard is not working correctly with TF < 1.4.
RiseML v1.0.3 (2018/01/29)
This release improves autoscaling on AWS using our installer.
RiseML v1.0.2 (2018/01/17)
This release brings persistence for the NFS provisioner. Together with the latest version of RiseML's AWS installer, you will get a persisted and auto-scaling RiseML installation on AWS!
- Configurable persistence for NFS provisioner
- Auto-scaling and persistence when using RiseML AWS installer (see website)
RiseML v1.0.1 (2017/12/20)
This release fixes a minor installation issue with Kubernetes Version 1.9
- Support installation on Kubernetes 1.9
- Minimum Kubernetes Version is now 1.8
RiseML v1.0.0 (2017/12/19)
This marks our 1.0.0 release
This tighter integration allows using any of the storage types that are supported in Kubernetes.
Second, we now provide a quick installer, which spins up a complete RiseML cluster (including GPU nodes) on AWS for you.
All you need is an AWS account and you're good to go in 15 minutes!
- Reworked storage integration with Kubernetes
- Quick installer for AWS
- Improved error reporting
- Enable support for different plans: basic, professional
- Fixed Tensorboard links with latest Tensorflow versions
- Many minor UX improvements (messages, command flags, ...)
Upgrading from Version v0.8.0
- backup your existing data volumes
- make sure to prepare
outputKubernetes volumes as described in our documentation
- remove the options
outputfrom you installation configuration (usually
- verify the configuration options in
riseml-config.ymlare still valid since they will be applied again (e.g., if you changed the account key to a different one it will be set to the one in
To update to the newest version:
helm update riseml helm upgrade riseml riseml-charts/riseml -f riseml-config.yml
If you need support with upgrading, please contact us.
RiseML v0.8.0 (2017/11/13)
To install RiseML, follow the installation guide.
- Introducing new features:
- TensorFlow and TensorBoard support
- Run distributed training experiments with TensorFlow
- Run hyperparameter experiments
- User Management
- Monitor experiments' CPU, RAM, and GPU stats
riseml.ymlsyntax for addressing frameworks
- Improved scalability of log streaming
dockerBuildinstallation parameter to
Upgrading from release candidates
There is no migration path from release candidates. To upgrade, you need to completely uninstall and re-install RiseML:
- uninstall via
helm del --purge riseml
- delete existing files on your database, git and log storage (only if persistence was configured)
- re-install according to installation instructions.