PyTorch
Description
PyTorch is a python package that provides two high-level features:
- Tensor computation (like Numpy) with strong GPU acceleration
- Deep Neural Networks built on a tape-based autograd system
Usually one uses PyTorch either as:
- A replacement for numpy to use the power of GPUs.
- a deep learning research platform that provides maximum flexibility and speed
Usage
Upcoming modulesystem change alert!
Due to large number of applications and their versions it is not practical to keep them explicitly listed at our wiki pages. Therefore an upgrade of modulefiles is underway. A feature of this upgrade will be the existence of default module for every application. This default choice does not need version number and it will load some (usually latest) version.
You can test the new version now by adding a line
source /cvmfs/software.metacentrum.cz/modulefiles/5.1.0/loadmodules
to your script before loading a module. Then, you can list all versions of pytorch and load default version of pytorch as
module avail pytorch/ # list available modules module load pytorch # load (default) module
If you wish to keep up to the current system, it is still possible. Simply list all modules by
module avail pytorch
and choose explicit version you want to use.
PyTorch as a Singularity container
Newer versions of PyTorch are available solely as Singularity images optimized for usage with NVidia GPUs (NVidia GPU Cloud, NGC). The NGC packages are placed in the directory /cvmfs/singularity.metacentrum.cz; you have to list the directory first to see its contents:
ls /cvmfs/singularity.metacentrum.cz
To use a selected version of PyTorch image, run the image within interactive job as:
qsub -I -l select=1:mem=16gb:scratch_local=10gb:ngpus=1:gpu_cap=cuda60:cuda_version=11.0 -q gpu -l walltime=4:00:00 singularity shell --nv /cvmfs/singularity.metacentrum.cz/NGC/PyTorch\:21.03-py3.SIF
More about Nvidia GPU cloud usage can be found at NVidia deep learning frameworks wiki page.
PyTorch as a modulefile
Older versions of PyTorch can be obtained also as modulefiles. Please keep in mind that the usage of PyTorch as modulefile is deprecated and newer versions will not be installed.
All currently installed versions of PyTorch support following methods of computations:
PyTorch 1.1.0:
pytorch-1.1.0_python-3.6.2_cuda-10.1(May 2019); version1.1.0- with Python 3.6.2, CUDA 10.1 and CuDNN 7.0
PyTorch 0.3.0:
pytorch-0.3.0_python-3.6.2_cuda-8.0(January 2018); version0.3.0.post4- with Python 3.6.2, CUDA 8.0 and CuDNN 7.0pytorch-0.3.0_python-2.7.6_cuda-8.0(January 2018); version0.3.0.post4- with Python 2.7.6, CUDA 8.0 and CuDNN 7.0
Usage:
(JESSIE)leontovyc_roman@tarkil:~$ module add pytorch-1.1.0_python-3.6.2_cuda-10.1
(JESSIE)leontovyc_roman@tarkil:~$ python -c 'import torch; print(torch.__version__);'
1.1.0
Documentation
You can find documentation at the official website: http://pytorch.org/docs/0.3.0/.
Official website
NVidia PyTorch containers: