ML Runtimes What's New
Major features and updates for ML Runtimes.
This release is available with ML Runtimes version 2023.05.1.
- This release of ML Runtimes includes the following features that improve support for
Large Language (LLMs) within CML:
- Git Large File Storage (LFS) support is now included in all ML Runtimes images.
Users can now easily download Large Language models via Git.
LFS replaces large files such as audio samples, videos, datasets, and graphics with text pointers inside Git, while storing the file contents on a remote server like GitHub.com or GitHub Enterprise.
- Upgraded CUDA to version 11.8.0 in JupyterLab Python CUDA runtimes. This change allows customers to use the latest pytorch version in CML. Affected Python versions are 3.7, 3.8, 3.9, and 3.10.
- NVIDIA GPU versions of PBJ Runtimes are now available for Python 3.7, 3.8, 3.9 and
3.10. The base image used for CUDA ML Runtimes are the “devel” series CUDA image
provided by NVIDIA. This change, for example, enables the usage of the Numba JIT
This change also supports the latest version of tensorflow in CML and some other libraries that need to compile GPU code in run-time.
- wget is now installed on all ML Runtimes. This enables users another way to download large language models in CML in addition to git-lfs.
- Git Large File Storage (LFS) support is now included in all ML Runtimes images. Users can now easily download Large Language models via Git.
- This release introduces a new Technical Preview Runtime, adding support for Anaconda, based on the JupyterLab editor. Conda environments can be created, modified, and activated from the Terminal without any limitations. For more information, see Using Conda Runtimes.
- New Python 3.10 Runtimes are available.
- Added the following extensions to our JupyterLab Runtimes (Python 3.8, 3.9, 3.10): Language Server Protocol (jupyterlab-lsp) and git (jupyterlab-git). Language Server Protocol integration provides coding assistance for JupyterLab (code navigation, hover suggestions, linters, autocompletion, rename).
- Added the git (jupyterlab-git) extension to our JupyterLab Runtimes (Python 3.7).
- The Workbench Runtimes is being deprecated in favor of PBJ. As a result, the CUDA version for the Workbench Runtimes is not updated for the current release.
- Implya client is now updated to version 0.18.0 which has support for authentication via JWT token.
- DSE-26724 wget is now installed on all ML Runtimes. This enables users to download large language models in CML two ways: via Gi and via wget.
- DSE-25078 The OS package
cmakeis now installed for all ML Runtime images. This change enables users to install packages like
- DSE-23762 Updated installed R package versions to make sparklyr R package usable in the same session as it was installed in.
- DSE-23751 Upgraded Jupyter to version 3.6.0 in all JupyterLab Runtimes.
- DSE-23613 Upgraded CUDA to version 11.8.0 in JupyterLab Python CUDA runtimes. This change was needed to allow customers to use the latest pytorch version in CML. Affected Python versions are 3.7, 3.8, 3.9, and 3.10.
- DSE-22425 Added OS packages to R Runtimes to support recent versions of the "devtools" package.
- DSE-20234 Jupyter session-specific data is no longer persisted in the Project directory. This prevents errors when starting multiple Jupyter sessions.