Training IBM PowerAI R5 samples with IBM Spectrum Conductor Deep Learning Impact 1.1
Helena. 270005V76W Visits (7703)
Authors: Sum Huynh
Published on December 22, 2017 / Updated on January 3, 2018
IBM Spectrum Conductor Deep Learning Impact 1.1 supports IBM
To start training PowerAI R5 samples with IBM Spectrum Conductor Deep Learning 1.1, first ensure the following:
This example assumes that PowerAI is installed under the default directory /opt/DL and that cifar10 is the sample model being trained.
Step 1: Log in to the management host as root.
Step 2: Two Caffe options are available in your PowerAI installation directory, caffe-ibm and caffe-bvlc. For the purposes of this example, activate caffe-ibm as follows:
# source /opt
Step 3: Run the get_cifar10.sh script to download the cifar10 dataset.
# cd /opt
Step 4: Run the create_cifar10.sh script to convert the cifar10 dataset to an LMDB for Caffe.
# cd /opt/DL/caffe-ibm/
Step 5: After running the conversion script, check that the following files and directories are created: cifar10_train_lmdb, cifar10_test_lmdb, and mean.binaryproto.
Step 6: Change the permissions on the newly created directories so that the LMDBs are readable, writable and executable by any users trying this sample.
# chmod –R 777 exam
Step 7: Create soft links from the original Caffe files to files that use the naming convention required by IBM Spectrum Conductor Deep Learning Impact.
# cd examples/cifar10
# ln -s cifa
# ln -sci
Step 8: Update the model for distributed training with IBM Fabric. Edit the cifa
Step 9: Use the IBM Spectrum Conductor Deep Learning Impact GUI to create the cifar10 dataset. Since this is a Caffe model, you need to create an LMDB dataset. For this example, the train and test folder should be set to /opt
Step 10: Use the IBM Spectrum Conductor Deep Learning Impact GUI to create cifar10 model. Since this is a Caffe model, you need to specify Caffe as the framework. For this example, the “Path” field in the GUI should be set to /opt
When specifying values for the cifar10 model, make sure that you:
More information on creating a training model can be found at http
Step 11: You can now start training and monitoring training progress.
To start training IBM PowerAI samples today, down
For more information about IBM Spectrum Conductor Deep Learning Impact and IBM PowerAI support, go to: www.