Skip to content

Release 2021.07.19
Compare
Choose a tag to compare
@harshad16 harshad16 released this 19 Jul 19:39
· 1616 commits to master since this release
v2021.07.19
7a231b5
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.

we have completed the release of 2021.07.19 🎉 🎊 🥳

Features

  • TensorFlow updates in recommendations

We have incorporated few additions to the recommender system when it comes to recent TensorFlow releases. Users of Thoth can now benefit from recommendations that take into account API symbols available also in the upcoming tensorflow~=2.6.0 release. The recommender system automatically suggests using the appropriate TensorFlow release based on API symbols used in the application (that matches API symbols available in the recommended TensorFlow release). The knowledge base was also extended with supported CUDA versions for TensorFlow 2.5 and upcoming TensorFlow 2.6.

Thoth's knowledge was extended with information that states which Python packages are packaged as RPMs in Fedora 34 and UBI8. When asking for advice, the recommender system notifies about Python packages available as RPMs in justifications in the respective distributions.

Number of Python packages packaged in Fedora 34 as RPMs: 2244
Number of Python packages packaged in UBI as RPMs: 62

With recent changes, we provide a link to PyPI's release page for each Python package installed from PyPI. This helps users to directly navigate to project information presented on PyPI.

The declarative interface to the resolution process now allows to state how the resolution process should look like when considering also Python packages already present in the runtime environment (container image). An example can be an adjusted resolution process when JupyterLab is installed in a specific version in the runtime environment (container image).

As we provide base container images with a prepared and tested set of packages ("predictive stacks") that are shipped with the container image, these container images can be additionally used as a base for installing packages as per user requests.

An example scenario can be a tested TensorFlow container image, that could be lacking TensorBoard package. If a user wants to install TensorBoard, the resolution process automatically detects TensorFlow release available in the container image and suggests a compatible TensorBoard release that will work with the preinstalled TensorFlow package shipped in the base image.

The recommendation engine now provides links to release notes pages available on GitHub (if any) for packages that are installed. For example, users will be pointed to numpy==1.21.0 release page if the given numpy is installed in the specific version. This information was aggregated for top 5000 Python packages on PyPI (top based on number of downloads). If you are a maintainer of a repository that provides release information on GitHub and your package is not listed in Thoth's database, feel free to add it to the thoth-station/prescriptions repository following the docs.

To make sure Python application developers use Python packages that are still actively developed, we warn when users use Python packages that have archived repositories on GitHub. This detection is performed for the top 5000 used packages on PyPI (based on the number of downloads).

We inform users if they consume packages from Operate First Pulp instance. The recommender system prints this information in the provided justification together with the recommended stack.

Component Updates

Thanks for the amazing work everyone. 💯

Assets 2
Loading