GitLab

As we discussed, this is handled here directly 
cc @thomwolf

Improve setup and requirements

Use -e only in docs targeted at contributors.

If a user copy-pastes  command line with [--editable], they will hit
an error. If they don't know the --editable option, we're giving them
a choice to make before they can move forwards, but this isn't a choice
they need to make right now.

Improve handling of optional imports

This is really just for consistency.

Else, the next line fails with a confusion exception because the spm
variable isn't defined.

ImportError isn't really appropriate when there's no import involved.

Most module use the simpler `raise` version. Normalize those that don't.

Most users don't need it.

This will improve version coverage.

If a user or contributor ran `pip install -e .` on transformers < 3.0,
pip created a transformers.egg-info directory next to the transformers
directory at the root of the repository.

In transformers 3.0, the source is in a `src` subdirectory.
`pip install -e .` creates a transformers.egg-info directory there.
However, pip will still pick transformers.egg-info from the previous
location. This is a bug: https://github.com/pypa/pip/issues/5466

Users and contributors are likely to hit this problem because the
documentation for transformers 3.0 relies heavily on extra_requires
which didn't exist in earlier versions, so aren't defined in a stale
transformers.egg-info directory.

If such a directory exists, remove it. It's autogenerated, gitignored
and not supposed to contain anything of value.

Also provide shortcuts in a Makefile.

It's superseded by the "docs" extras.

It's redundant with setup.py and, also, incomplete (e.g. numpy).

It was generated once, likely in a non-reproducible way (pip freeze
in a contributor's local environment), and never updated.

Take advantage of this to simplify the Circle CI configuration.

Don't bother with tensorboardX: it's a fallback for PyTorch < 1.1.0.

Every other file uses this pattern.

requirements.txt isn't up to date.

This should fix API docs, which went AWOL with yesterday's changes.

Fix doc link in README

fix error due to wrong argument name to Tensor.scatter()

Remove support for Python 2

On Python 3, `open is io.open`.