ML Runtimes What's New

Major features and updates for ML Runtimes.

Version 2021.02

New features

  • Nvidia GPU support - Runtimes supporting Nvidia GPUs are available with preinstalled CUDA software for both Workbench and JupyterLab editors. (See ML Runtimes Nvidia GPU Edition.)
  • Runtimes supporting R 3.6 and R 4.0 available for Workbench editor
  • Support for JupyterLab runtimes - The JupyterLab runtimes now use JupyterLab 3.0 (was 2.2) and are considered generally available using Standard Edition ML Runtimes. See JupyterLab blog for notes on this upgrade.
  • Python runtimes include a C++ compiler - Python runtimes now include a C++ compiler (g++), which enables the installation and use of certain Python libraries such as impyla and pystan.

    Cloudera bug: DSE-14492

  • Preinstalled Python libraries (See ML Runtimes 2021.02) are now installed in /usr/local/lib rather than /var/lib/cdsw, which is where they were installed in engine:13 and runtimes 2020.11 (DSE-12177). This means that you can upgrade these packages more easily, but there are some packages you should not upgrade. (See Upgrading R and Python Packages)
  • ML runtimes are supported on CDSW (version 1.9.0 or later).

Fixed issues

  • DSE-13743 - Idle JupyterLab sessions are now ended after around IDLE_MAXIMUM_MINUTES. (See ML Runtimes Environment Variables Environment Variables).
  • DSE-14979 - matplotlib figures restore the styling used in engine:13 (MLRuntimes 2020.11 used the matplotlib defaults)
  • DSE-12881 - For Python runtimes, py4j is now installed.
  • Security fixes, python library version updates