1.5.4 iFix installation

mldl-repo-local update method

To upgrade systems where PowerAI Base is already installed:

Note: For PowerAI Enterprise    

• Update the mldl-repo-local package on each system of the cluster (Step 2)
• For each conda environment, make sure that no training's are running before updating the PowerAI framework packages (Step 3)
• Update conda environments for dlipy2 and dlipy3 (Step 4)

1. Download the 201912090734 version of the mldl-repo-local package

2. Update the mldl-repo-local package on the system

   • sudo rpm -Uhv mldl-repo-local-5.4.0-201912090734.9ecac53.ppc64le.rpm

3. Update the PowerAI framework packages

   All systems:

   • sudo yum update caffe-bvlc caffe-ibm mldl-conda-repo pytorch tensorboard tensorflow

   Systems with python 3 support (above update plus):

   • sudo yum update caffe-bvlc-py3 caffe-ibm-py3 pytorch-py3 tensorboard-py3 tensorflow-py3

4. Update existing conda environments

   First update the conda install to at least version 4.6.12:

   • conda install conda=4.6.12

   Then, for each existing conda environment:

   • Activate the environment

     • conda activate ...

     • For each installed PowerAI Base framework:

       Update the framework's pre-requisites:

       • /opt/DL/<framework>/install-dependencies

     • Update numpy to at least 1.16 only in Python 3.6 environments without snap-ml installed:

       • conda install numpy=1.16

     • Avoid broken version (3.0.0) of matplotlib

       • Python 2.7 env: conda install matplotlib=2.2.3
       • Python 3.6 env: conda install matplotlib=3.0.1

   • Deactivate the environment

     • conda deactivate ...

mldl-repo-local new install method

Follow existing KC installation instructions, just starting with the 201912090734 version of the mldl-repo-local local RPM package.

https://www.ibm.com/support/knowledgecenter/SS5SF7_1.5.4/navigation/pai_install.html