Supported deep learning frameworks

IBM Spectrum Conductor Deep Learning Impact 1.1.0supports different deep learning frameworks depending on your hardware. At least one supported framework must be installed.

If you are installing IBM Spectrum Conductor Deep Learning Impact for Linux (64-bit), use one of the following deep learning frameworks.
Table 1. Deep learning frameworks that work with IBM Spectrum Conductor Deep Learning Impact 1.1.0 for Linux (64-bit)
Framework Version Training options
TensorFlow 1.1
  • Single node training
  • Distributed training with TensorFlow
  • Distributed training with IBM Fabric
Caffe 1.0
  • Single node training
  • Distributed training with IBM Fabric
Note: IBM Spectrum Conductor Deep Learning Impact on Linux for POWER LE (64-bit) with IBM PowerAI 1.5.2 is only available with IBM PowerAI Enterprise 1.1. For more information, see Release notes for IBM Spectrum Conductor Deep Learning Impact 1.1.0.
If you are installing IBM Spectrum Conductor Deep Learning Impact 1.1.0 for Linux for POWER LE (64-bit), only the frameworks included with IBM PowerAI 1.5.0 are supported, including the following.
Table 2. Deep learning frameworks that are supported for IBM Spectrum Conductor Deep Learning Impact 1.1.0 for Linux for POWER LE (64-bit)
PowerAI Framework Version Training options
1.5.0 TensorFlow 1.4
  • Single node training
  • Distributed training with TensorFlow
  • Distributed training with IBM Fabric
  • Distributed training with IBM Fabric and auto-scaling
IBM Caffe 1.0
  • Single node training
  • Distributed training with IBM Fabric
  • Distributed training with IBM Fabric and auto-scaling
Caffe 1.0
  • Single node training
  • Distributed training with IBM Fabric
  • Distributed training with IBM Fabric and auto-scaling
1.5.1 TensorFlow 1.5
  • Single node training
  • Distributed training with TensorFlow
  • Distributed training with IBM Fabric
  • Distributed training with IBM Fabric and auto-scaling
IBM Caffe 1.0
  • Single node training
  • Distributed training with IBM Fabric
  • Distributed training with IBM Fabric and auto-scaling
Caffe 1.0
  • Single node training
  • Distributed training with IBM Fabric
  • Distributed training with IBM Fabric and auto-scaling
Note:
  • IBM Spectrum Conductor Deep Learning Impact installation requires at least one supported framework to be installed. However, it is recommended that you install both TensorFlow and Caffe or IBM Caffe. If either framework is missing, the option for the missing framework will not work in the cluster management console. If both Caffe and IBM Caffe are installed, IBM Spectrum Conductor Deep Learning Impact uses IBM Caffe by default.
  • To use the Distributed training with Caffe training engine, install CaffeOnSpark 1.6.1, 2.1.1 or 2.2.0.