Natural language processing

https://github.com/huggingface/transformers/blob/546dc24e0883e5e9f5eb06ec8060e3e6ccc5f6d7/src/transformers/models/gpt2/modeling_gpt2.py#L698

Assertions can't be relied upon for control flow because they can be disabled, as per the following:

$ python --help
usage: python [option] ... [-c cmd | -m mod | file | -] [arg] ...
...
-O     : remove assert and __debug__-dependent statem

In gensim/models/fasttext.py:

    model = FastText(
        vector_size=m.dim,
        vector_size=m.dim,
        window=m.ws,
        window=m.ws,
        epochs=m.epoch,
        epochs=m.epoch,
        negative=m.neg,
        negative=m.neg,
        # FIXME: these next 2 lines read in unsupported FB FT modes (loss=3 softmax or loss=4 onevsall,
        # or model=3 supervi

Is your feature request related to a problem? Please describe.
I want to evaluate multiple datasets (same formatting, they can share the same dataset reader). The "evaluate" command takes much longer to load the model than to evaluate.

Describe the solution you'd like
support passing multiple input files and output files to the "evaluate" command

**Describe alternatives you've cons

Running pytest with the new Python 3.9.4 shows a number of deprecation warnings about future failures to anticipate:

============================= test session starts ==============================
platform linux -- Python 3.9.4, pytest-6.2.0, py-1.10.0, pluggy-0.13.1

[....]
=============================== warnings summary ===============================
chunk.doctest::chunk.doctest
<d

Hi
Wikiann dataset needs to have "spans" columns, which is necessary to be able to use this dataset, but this column is missing from huggingface datasets, could you please have a look? thank you @lhoestq

I encountered a problem during benchmarking when running Flows on one machine with a high number of shards (64 in my case).
In that case there are sometimes port conflicts, because the Flow assigns the same port to multiple shards, which does not work on the same machine.
Of course that behaviour would be fine if all of those shards would live on different machines, which should be the case for