Colab compatibility

Google updates Colab’s runtime (Python, PyTorch, fastai, …) every few weeks. This package is built and tested against that runtime, not against “whatever pip picks”.

Checking your runtime

cu.env.check()

Runtime: Google Colab, Python 3.13.15 (tested: 3.13.15, Colab gpu runtime synced 2026-10-12)
package        source                         installed     tested        status
torch          provided by Colab              2.11.0+cu130  2.11.0+cu130  ok
fastai         provided by Colab              2.8.12        2.8.12        ok
...
torchinfo      installed by colabutilscs4337  1.8.0         -             ok

different usually means Colab updated since this release, or the notebook is pinned to an older runtime (Runtime > Change runtime type > Runtime version). Things normally still work; if not, upgrade the package and restart the session.

How compatibility is maintained

  • Nothing heavy is a dependency. PyTorch, torchvision, fastai, NumPy, and Matplotlib are used from Colab, so installing this package can never replace them.

  • Tested inside Colab’s own image. Every change runs the tests inside Google’s published Colab runtime image (us-docker.pkg.dev/colab-images/public/cpu-runtime) and checks that pip install "colabutilscs4337[all]" changes no preinstalled package.

  • Tested with Colab’s exact versions. A second job installs the versions listed in googlecolab/backend-info on Python 3.12 and 3.13.

  • Weekly drift check. A scheduled job re-runs everything against the newest Colab image and opens a pull request when Google’s published versions change.

  • Manual GPU check before each release with notebooks/smoke_test.ipynb.