[#38] upgrade to PyTorch 1.7.1
this does a few things:
-
adds separate ci/environment-*.yml conda environment files specifically for CI, which pin versions more strictly
- In particular, the environment-cuda.yml installs our custom build of PyTorch which is needed in order to be able to use it on kepler.lri.fr (where we're running our builds) which has a (slightly) older GPU architecture version no longer supported by the official PyTorch builds
-
misc minor fixes (one originally from !32 (closed)) and another fix needed for newer Pandas versions, as well as some workarounds to DeprecationWarnings emanating from dependencies of dependencies