Fix run_squad.py to work with BART

[tomgrek]

BART model should be among the set that doesn't use token_type_ids.

[codecov]

Codecov Report

Merging #6756 into master will increase coverage by 0.16%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #6756      +/-   ##
==========================================
+ Coverage   79.65%   79.82%   +0.16%     
==========================================
  Files         157      157              
  Lines       28479    28479              
==========================================
+ Hits        22686    22734      +48     
+ Misses       5793     5745      -48     

Impacted Files	Coverage Δ
src/transformers/modeling_tf_xlm.py	18.94% <0.00%> (-74.32%)	⬇️
src/transformers/modeling_roberta.py	77.37% <0.00%> (-19.71%)	⬇️
src/transformers/file_utils.py	82.66% <0.00%> (+0.25%)	⬆️
src/transformers/tokenization_utils.py	89.84% <0.00%> (+0.39%)	⬆️
src/transformers/tokenization_bert.py	91.51% <0.00%> (+0.44%)	⬆️
src/transformers/tokenization_utils_fast.py	94.28% <0.00%> (+1.42%)	⬆️
src/transformers/tokenization_openai.py	84.09% <0.00%> (+1.51%)	⬆️
src/transformers/generation_tf_utils.py	86.46% <0.00%> (+3.00%)	⬆️
src/transformers/tokenization_transfo_xl.py	42.48% <0.00%> (+3.75%)	⬆️
src/transformers/tokenization_utils_base.py	93.76% <0.00%> (+7.18%)	⬆️
... and 2 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 61b9ed8...7ba95ec. Read the comment docs.

[LysandreJik]

Nice, thanks!