Difference between revisions of "Infrastructure/software/pytorch"

From Nordic Language Processing Laboratory
Jump to: navigation, search
(Available Versions)
(Usage)
 
Line 22: Line 22:
  
 
<pre>
 
<pre>
module purge
 
 
module use -a /proj*/nlpl/software/modulefiles
 
module use -a /proj*/nlpl/software/modulefiles
 
module load nlpl-pytorch
 
module load nlpl-pytorch
 
</pre>
 
</pre>
  
There is a short sample program that test availability of cpu vs.
+
There is a short sample program that tests availability of cpu vs.
 
gpu computing devices.
 
gpu computing devices.
  

Latest revision as of 12:33, 14 November 2018

Background

In mid-2018 at least, PyTorch is one of the widely used programming frameworks for Deep Learning (with corporate backing by facebook); in and of itself PyTorch is not NLP-specific, but it is a critical dependency for several of the NLPL strands of activity.

Usage

The module nlpl-pytorch provides a PyTorch installation in a Python 3.5 virtual environment. Some general background on ‘mixing and matching’ of NLPL Python modules is discussed as a separate page. Besides PyTorch and its dependencies (e.g. NumPy), the virtual environment includes a selection of popular add-on packages, e.g. SciKit-Learn, the Python Data Analysis Library (Pandas), and Keras. This installation should support both cpu and gpu nodes on Abel and Taito.

module use -a /proj*/nlpl/software/modulefiles
module load nlpl-pytorch

There is a short sample program that tests availability of cpu vs. gpu computing devices.

python /proj*/nlpl/software/pytorch/0.4.1/test.py

Available Versions

On Abel and Taito, as of September 2018, PyTorch 0.4.1 is available and the default version of the nlpl-pytorch module.

Installation on Abel

module purge
module load cuda/8.0
module load python3/3.5.0
cd /projects/nlpl/software
mkdir pytorch
virtualenv pytorch/0.4.1

Next, we need to create a module definition, in this case /projects/nlpl/software/modulefiles/nlpl-pytorch/0.4.1.

module load nlpl-pytorch/0.4.1
pip install --upgrade pip
pip install --upgrade numpy pillow six
pip install torch torchvision
wget -O /tmp/requirements.txt \
  https://raw.githubusercontent.com/OpenNMT/OpenNMT-py/master/requirements.txt
pip install -r /tmp/requirements.txt

Installation on Taito

module purge
module load cuda/9.0
module load python-env/3.5.3
cd /proj/nlpl/software
svn co http://svn.nlpl.eu/software/pytorch
virtualenv pytorch/0.4.1

Next, we need to create a module definition, in this case /proj/nlpl/software/modulefiles/nlpl-pytorch/0.4.1.lua.

module load nlpl-pytorch/0.4.1
pip install --upgrade pip
pip install --upgrade -r /proj/nlpl/software/pytorch/0.4.1/modules.txt