[NLPL Task Force (A)] PyTorch and TensorFlow on Abel

[Stephan Oepen]

colleagues,

with a bit of creativity, we have put PyTorch and TensorFlow
installations on Abel under the NLPL umbrella.  these installations
support both cpu and gpu nodes and are ‘native’ in the sense of not
requiring containerization.  thus, they should play nicely with all
other software installed on Abel through the module infrastructure, as
well as with user-maintained software installed locally into your home
directory.

we plan on installing additional NLP-specific software this fall,
which at least in part will build on PyTorch and TensorFlow (e.g.
OpenNMT).  hence, it would be good to have these installations tested
somewhat more broadly.  please experiment and do not hesitate to send
problem reports, suggestions for improvement, or other feedback to the
NLPL infrastructure task force: ‘infrastructure at nlpl.eu’.

for some background on these installation, please see:

http://wiki.nlpl.eu/index.php/Infrastructure/software/pytorch
http://wiki.nlpl.eu/index.php/Infrastructure/software/tensorflow

regarding gpu availability on Abel, my impression is things are not at
all bad lately.  but more gpus will be purchased and installed before
the end of the year, both at UiO and at the national level in norway.

happy computing!  oe