What's included

WML Accelerator includes these components and frameworks.

Components
Deep learning frameworks
Technology preview packages

Components

The following components are included with WML Accelerator:

IBM Spectrum Conductor™ Deep Learning Impact version 1.2.4
For information about this component, see these topics: Working with IBM Spectrum Conductor Deep Learning Impact.
Elastic distributed inference version 1.2.5
IBM Spectrum Conductor version 2.4.1:
WML CE version 1.7.0
- IBM Watson Machine Learning Community Edition Knowledge Center
- Getting started with frameworks

Deep learning frameworks

The following frameworks are included with WML Accelerator for use with IBM Spectrum Conductor Deep Learning Impact. Some frameworks are available via the user interface. Others can only be used through the command line interface. For details, see Using frameworks via command line interface

Table 1. Included frameworks
Framework	Version	Python 3.7 support³	Python 3.6 support⁴	Supported via user interface	Supported via command line¹	Training options
TensorFlow	2.1	Yes	Yes	X	X	Single node training Distributed training
Keras	2.3.0	Yes	Yes		X	Single node training Distributed training Elastic distributed training
IBM® Caffe	1.0		No	X	X	Single node training
PyTorch	1.3.1		Yes	X	X	Single node training Elastic distributed training
snap.ml	1.6.0	No	Yes	X²	X	Single node training
Important: For information on framework support via command line, see Using frameworks via command line interface for details. snap.ml is available as a notebook only. By default, frameworks that use Python 3.7 are installed to conda environment dlipy37-wmlce170. By default, frameworks that use Python 3.6 are installed to conda environment dlipy36-wmlce170. If you upgraded from WML Accelerator 1.2.1 to 1.2.2, some WML CE 1.6.1 frameworks are still supported. By default, these frameworks are installed to conda environment dlipy36-wmlce161. In this case, it is recommended that you update these models to work with WML CE 1.7.0. Framework support for models created with WML CE 1.6.1 includes: TensorFlow 1.14.0 is supported via command line and user interface. TeamKeras 2.2.4 for elastic distributed training is supported via command line.

Technology preview packages

The Program includes the following technology preview packages:

TFLMSv1
SnapML: snap-ml-spark library
pai4sk: Decision Tree and Random Forest algorithms from SnapML
RAPIDS.AI: cuML and cuDF

Before using technology preview packages, read the information in Technology Preview Code.