MLReef is currently in closed early access. Thing will change on a day to day basis!

This will guide you through the most significant changes in MLReef. We will try to keep it updated as much as we can.

Release log

r-05/2022 - MLReef v.1.0.0

This release brings MLReef to the first stable version of 1.0.0. The release contains a lot of changes and fixes, documentation updates and UI/UX improvements.

The major changes are these:

  • Refurbished publishing process of AI modules:
    • No need for parameter annotations. Simply include a command-line interface, such as argparse to your scripts. We will automatically parse these to make your parameters addressable in the pipeline creator.
    • You can now review your parameters during publishing, as we parse them prio of running the publishing pipeline.
    • You now can select multiple base environments, including an ANACONDA environment and a JVM environment.
    • You can now manage your dependencies with anaconda (miniconda) during publishing by selecting the miniconda base environment.
  • Several changes to the registration process. These include the necessity to provide an organization.
  • Changes to the home screen. Now, you can access all your organization´s projects here.
  • You can manage your members of your organization in your home screen. Give them permission levels and access rights as you see fit.
  • MLReef now supports data connectors. The first data connector is Amazon AWS S3 buckets. The configuration is simple and all data stored in your S3 bucket is displayed in your ML Project in your data tab.
  • Several changes to this documentation.


With the latest release we solved many bugs and issues during this beta phase and worked on general stability.

In addition, now you can:

  • Create multiple experiments at the time (although there is still no concurrent pipelines)
  • General stability and improved test coverage
  • Rework of the publishing process to be more light-weight and agile
    • Including publishing up to 10 versions of master and unlimited branches
  • Rework of the experiment tracking and experiment management section now including graphs and more functionality
  • Include the possibility to see the entry point script in all pipelines


This release moves MLReef to beta stage, involving broad changes to the entire infrastructure and baseline.

Now, you can:

  • Create code projects (repositories)
    • Publish code repositories
    • Re-publish code repositories via the publishing wizard or via commit to master of your code repository master branch
    • Published code projects are stored in a docker registry and explorable / usable in the data pipelines directly
  • General stability and improved test coverage
  • Many smaller features, such as creating new files
  • Ad-hoc visualizations based on tabular data (insights/graphs)
  • Move to kubernetes cluster (for hosted version)
  • Development of "Nautilus", an offline and on-premise version of MLReef


In this release we included:

  • Create ML projects
  • Use basic repository functions, such as creating branches, merging and reviewing MR.
  • Upload data into your ML project
  • Create basic experiments
    • You can only use one experiment at the time. Currently only ResNet 50 Model (images) and a dummy model are included.
    • Training log
    • Download experiment artifacts (such as the model binary and metric values)
    • Visualize experiment details with params and data source.
  • Create basic data processing pipeline (DataOps)
    • You can create data pipelines using 3 available data operations for images
    • Concatenate data processors in one pipeline (order is kept)
    • Creates Dataset (currently hold as a separate branch - you can merge it into your data repository)
  • Create basic data visualization using currently only one operator (t-SNE)
  • Infrastructure supports GPU and CPU execution