Reduce the costs of Machine Learning projects with Preemptible Instances on GCP
Spell makes it easy to set up and manage Google Cloud Platform Compute (GCP) Preemptible Instances. A Preemptible Instance can save you up to 80% off regular instances, and with Spell Auto Backups, you never have to worry about losing your work if the instance terminates.
Preemptible instances take advantage of excess GCP Compute Engine capacity provided at a discounted rate, so their availability varies with usage, and they can terminate at any time. If your instance is terminated, Spell attaches the volume to a new CPU to finish saving all the files changed during your run. If your code makes use of checkpoints, you can save money by picking up your interrupted training where it left off.
Spell also makes it easy to set up and manage AWS Spot Instances. More details of how Spell simplifies and reduces the cost of machine learning projects with Preemptible and Spot Instances in our docs https://spell.ml/docs/ownvpc_machine_types/#available-instance-types
Keeping Projects Secure with Private Docker Images
Spell now supports private docker images on AWS and GCP. Teams and Enterprise users using Elastic Container Registry (ECR) on AWS or Google Container Registry (GCR) on GCP only need to add Spell permissions once, for the images to be accessible across all users and runs in their account.
To easily add private docker images in a registry from your cloud account, simply use the spell cluster add-docker-registry command. More details at https://spell.ml/docs/reference/#spell-cluster-add-docker-registry
Keep ML projects synced on Github within JupyterLab Workspace
Spell now makes it more efficient to keep your project files synced on Github. In JupyterLab Workspaces, the GitHub tab allows you to specify your repository and make pull and push changes with just one click saving you time from switching between different tools. This streamlines the workflow of individuals and teams collaborating on projects and works with private and public GitHub repositories. Learn more about JupyterLab Workspaces https://spell.ml/docs/workspaces_overview/#the-notebook-server
Compare the performance of multiple ML experiments with TensorBoard
TensorBoard provides a variety of features, including tracking metrics, visualizing model graphs, projecting embeddings to a lower-dimensional space, displaying images/text/audio data, and much more. Spell now makes it easy for you to use TensorBoard for multiple completed runs directly from the Runs page. On the Runs page in Spell’s web console, select multiple completed runs you would like to view using the checkboxes on the left, filter Runs configured with Tensorboard by clicking the 'T' icon in the filters bar, then. This allows you to more efficiently understand key metrics across different runs such as loss and how they change as training progresses.
We also support Tensorboard in JupyterLab Workspaces via the Tensorboard extension.
Learn more about using Tensorboard on Spell here https://spell.ml/docs/tensorboard