We’re excited to announce releases 0.14.6 through 0.15.1 of the Determined deep learning training platform. Here’s what we’ve been working on in April:

Determined now arrives as a single package, rather than being spread out over multiple packages for the CLI, deploy, and common. We encourage you to upgrade to the latest version - 0.15.1 - but there is a chance that the det command may break for some users due to pip limitations. If this happens, please uninstall the old packages and reinstall Determined from our single package. Please reference our documentation if you have issues with the upgrade.

PyTorch Lightning Adapter: PyTorch Lightning users can use the Lightning Adapter as a quick way to train models with key Determined features. This allows you to get the benefits of distributed training, GPU management, and mid-epoch preemption while using PyTorch Lightning. Check out our documentation to get started!

Hyperparameter Visualizations: During our last release wrap-up, we talked about the launch of our new hyperparameter visualization tools to better highlight the relationship between hyperparameters and model performance. In the lastest release, we have added support for showing categorical hyperparameters on the heat map and scatter plot visualizations.

Historical GPU Allocation: The Cluster page now provides historical allocation information for GPU hours daily. This data calculates total time, as well as breakdowns by user, label, and resource pool.

You can read the latest release notes here. To get started with Determined check out the install guide and the associated CLI.

If you have questions, feedback, or comments, we welcome you to get in touch with us on our mailing list. Feel free to ping us on our Determined Community Slack as well!