What's new or changed in IBM Spectrum Conductor Deep Learning Impact 1.2.3

List of what's new or changed in IBM Spectrum Conductor Deep Learning Impact 1.2.3.

IBM Spectrum Conductor Deep Learning Impact 1.2.3 has the following key changes:
  • Installation and upgrade
  • Features and functionality
  • Deep learning frameworks
  • Samples

Installation and upgrade

You can now perform a rolling upgrade from IBM Spectrum Conductor Deep Learning Impact 1.2.2 to 1.2.3.
Note: If you are on IBM Spectrum Conductor Deep Learning Impact 1.2.1 or earlier and you want to upgrade to the latest version, contact IBM Support.
IBM Spectrum Conductor Deep Learning Impact 1.2.3 can be installed on Power and x86 systems. The following environment variables are new or changed:
  • New environment variable: DLI_REDIS_PASSWORD
  • Environment variable: METRICS_STREAMING was changed to DLI_METRICS_STREAMING_ENABLED
  • Default value for DLI_INSIGHTS_MONITOR_PORT changed from 5000 to 50000
  • Default value for DLI_INSIGHTS_OPTIMIZER_PORT changed from 5001 to 50001

Features and functionality

The following features and functionality is new or has changed in IBM Spectrum Conductor Deep Learning Impact:
  • The IBM Spectrum Conductor Deep Learning Impact RESTful API now supports additional hyperparameter optimization search parameters.
  • Previous setfacl permission dependencies have been removed.
  • User datasets and models are now stored under DLI_RESULT_FS. Previously they were stored in DLI_SHARED_FS.

Deep learning frameworks

IBM Spectrum Conductor Deep Learning Impact 1.2.3 now supports TensorFlow 1.14 and continues to support IBM Caffe 1.0 and PyTorch 1.0.1. IBM Spectrum Conductor Deep Learning Impact no longer supports TensorFlow 1.13.1.

New training options in IBM Spectrum Conductor Deep Learning Impact 1.2.3, include:
  • Elastic distributed training for the Keras framework plugin

Samples

Deep learning model samples that are instrumented to work with IBM Spectrum Conductor Deep Learning Impact 1.2.3 are available from IBM Cloud. These samples come with deep learning capabilities already enabled. To learn more, see Sample models.