PyTorch¶

PyTorch is an open source machine learning library for Python, based on Torch. It is used for applications such as natural language processing.

About PyTorch on Bede¶

Note

GPU must be requested in order to enable GPU acceleration by adding the flag e.g. --gpus=1 to the scheduler command or job script. See Running and Scheduling Tasks on Bede for more information.

Bede has a locally installed IBM Watson Machine Learning Community Edition (WML CE) Anaconda channel that provides versions of Tensorflow, PyTorch and their dependencies especially built for the POWER architecture.

Installation in Home Directory¶

First request an interactive session, e.g. see Running and Scheduling Tasks on Bede.

Then PyTorch can be installed by the following

# Load the conda module
module load Anaconda3/2020.02

# Add the local WML CE channel to the conda search path
 conda config --prepend channels file:///opt/software/apps/ibm_wmlce/wmlce-1.7.0-mirror/

# Create an conda virtual environment called e.g. named 'pytorch'
conda create -n pytorch python=3.6

# Activate the 'pytorch' environment
source activate pytorch

# Install PyTorch
conda install torch

Every Session Afterwards and in Your Job Scripts

Every time you use a new session or within your job scripts, the modules must be loaded and conda must be activated again. Use the following command to activate the Conda environment with PyTorch installed:

# Load the conda module
module load Anaconda3/2020.02

# Activate the 'pytorch' environment
source activate pytorch

Testing your PyTorch installation¶

To ensure that PyTorch was installed correctly, we can verify the installation by running sample PyTorch code e.g. an example from the official PyTorch getting started guide (replicated below).

Here we construct a randomly-initialized tensor:

import torch
x = torch.rand(5, 3)
print(x)

The output should be something similar to:

tensor([[0.3380, 0.3845, 0.3217],
        [0.8337, 0.9050, 0.2650],
        [0.2979, 0.7141, 0.9069],
        [0.1449, 0.1132, 0.1375],
        [0.4675, 0.3947, 0.1426]])

Additionally, to check if your GPU driver and CUDA is enabled and accessible by PyTorch, run the following commands to return whether or not the CUDA driver is enabled:

import torch
torch.cuda.is_available()