Added Seq2Seq tasks

[SeanNaren]

What does this PR do?

Add Translation/Summarization and general Seq2Seq Task. This is a first pass, and will need more iteration in the future:

Painpoints:

• I didn't have enough time to train a mt5 model on translation, as this would be common for all translation tasks. This is due to requiring some additional pre-processing, and longer training times. As a stop gap, I've used a student model trained on the en/ro translation task as I think is the case in HF, but that shouldn't be what we deliver in the future.
• Upload WMT translation sets in different languages!

Left to do:

• Train Summarization model, ensure this works with predict
• Clean up Translation predict output, check single string input works
• Upload datasets/models to S3
• Create tests replicating classification tests
• In a separate PR (done by someone familiar with the doc structure if poss) we need to add these additional tasks

[codecov]

Codecov Report

Merging #37 (cdf5859) into master (d5409bd) will decrease coverage by 0.26%.
The diff coverage is 76.05%.

@@            Coverage Diff             @@
##           master      #37      +/-   ##
==========================================
- Coverage   81.17%   80.91%   -0.27%     
==========================================
  Files          32       46      +14     
  Lines        1036     1383     +347     
==========================================
+ Hits          841     1119     +278     
- Misses        195      264      +69     

Flag	Coverage Δ
unittests	80.91% <76.05%> (-0.27%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
flash/text/seq2seq/core/finetuning.py	35.71% <35.71%> (ø)
flash/text/seq2seq/summarization/metric.py	53.44% <53.44%> (ø)
flash/text/seq2seq/core/model.py	68.25% <68.25%> (ø)
flash/text/seq2seq/translation/model.py	72.22% <72.22%> (ø)
flash/core/trainer.py	82.50% <75.00%> (+10.00%)	⬆️
flash/text/seq2seq/summarization/utils.py	75.00% <75.00%> (ø)
flash/text/seq2seq/summarization/model.py	82.35% <82.35%> (ø)
flash/text/seq2seq/summarization/data.py	84.61% <84.61%> (ø)
flash/text/seq2seq/translation/data.py	84.61% <84.61%> (ø)
flash/text/seq2seq/core/data.py	85.33% <85.33%> (ø)
... and 26 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 d5409bd...23aacf9. Read the comment docs.

[pep8speaks]

Hello @SeanNaren! Thanks for updating this PR.

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-02-02 14:18:05 UTC

[Borda]

what is flash_notebooks/.lock for?
why the strange notebook changes?

go