Amazon SageMaker now supports a hosted TensorBoard experience
Today, AWS announces the general availability of Amazon SageMaker with TensorBoard, which provides a hosted TensorBoard experience. This launch allows you to use TensorBoard to visualize and debug model convergence issues for Amazon SageMaker training jobs.
Today, AWS announces the general availability of Amazon SageMaker with TensorBoard, which provides a hosted TensorBoard experience. This launch allows you to use TensorBoard to visualize and debug model convergence issues for Amazon SageMaker training jobs.
TensorBoard is an observability tool commonly used by data scientists to track model accuracy and log loss on training and validation sets. With this capability, data scientists can save development time by visualizing the model architecture to identify and remediate convergence issues, such as validation loss not converging or vanishing gradients. Further, the access and management of this capability is automated using Amazon SageMaker Python SDK. By providing TensorBoard as a hosted experience, data scientists will gain optimized S3 read access for TensorBoard log data and will not have to manually install and configure TensorBoard.
Amazon SageMaker with TensorBoard is available in the following regions: US East (Ohio), US East (N. Virginia), US West (Oregon), Europe (Frankfurt), and Europe (Ireland) using ml.r5.large instance types. We are providing SageMaker with TensorBoard for free for the next 2 months to all SageMaker customers. Please see here information on pricing that will apply following the 2 month period.
To learn more, see the documentation page.