add T5 fine-tuning notebook [Community notebooks]

[patil-suraj]

@patrickvonplaten
This is the second notebook which shows how to fine-tune T5 for multiple tasks with text-to-text approach (IMDB, emotion classification, SWAG) that we discussed in issue #4426 . I didn't find emotion and SWAG dataset in the nlp library so I decided to keep my original dataset code to keep everything unified.

Also there's growing interest in pytoch-lightning so I decided to keep the lightning trainer. But if you think I should use HF Trainer then I can add that as well.

[codecov-commenter]

Codecov Report

Merging #4462 into master will increase coverage by 0.04%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #4462      +/-   ##
==========================================
+ Coverage   77.93%   77.98%   +0.04%     
==========================================
  Files         123      123              
  Lines       20430    20430              
==========================================
+ Hits        15922    15932      +10     
+ Misses       4508     4498      -10     

Impacted Files	Coverage Δ
src/transformers/modeling_utils.py	90.41% <0.00%> (-0.12%)	⬇️
src/transformers/modeling_tf_utils.py	88.66% <0.00%> (+1.80%)	⬆️

---

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 8f1d047...ca0d2a0. Read the comment docs.

[patrickvonplaten]

That's awesome. Thanks @patil-suraj! @mariamabarham @lhoestq - might it be interesting to add emotion classification and swag to nlp?

[patrickvonplaten]

Reworded the description a bit - hope that's ok @patil-suraj

[patil-suraj]

Reworded the description a bit - hope that's ok @patil-suraj

@patrickvonplaten yes, it's more clear now. Thank you!