Tf trainer cleanup

[sgugger]

New version of #6015

Will harmonize the public customization hooks and document everything properly once this is merged.

[codecov]

Codecov Report

Merging #6143 into master will increase coverage by 1.02%.
The diff coverage is 25.00%.

@@            Coverage Diff             @@
##           master    #6143      +/-   ##
==========================================
+ Coverage   78.35%   79.38%   +1.02%     
==========================================
  Files         146      146              
  Lines       26403    26416      +13     
==========================================
+ Hits        20689    20970     +281     
+ Misses       5714     5446     -268     

Impacted Files	Coverage Δ
src/transformers/trainer_tf.py	16.14% <25.00%> (-0.41%)	⬇️
src/transformers/modeling_tf_gpt2.py	65.42% <0.00%> (-29.91%)	⬇️
src/transformers/tokenization_marian.py	68.14% <0.00%> (-25.67%)	⬇️
src/transformers/tokenization_xlnet.py	66.66% <0.00%> (-23.43%)	⬇️
src/transformers/file_utils.py	82.20% <0.00%> (-0.29%)	⬇️
src/transformers/generation_tf_utils.py	85.96% <0.00%> (-0.26%)	⬇️
src/transformers/modeling_tf_mobilebert.py	96.77% <0.00%> (+73.38%)	⬆️

---

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 54f9fbe...46b6cb4. Read the comment docs.

[LysandreJik]

Cool, this is clean. LGTM, love the warnings.

[jplu]

LGTM!