Correct order of overflowing tokens for LayoutLmV2 tokenizer

[Apoorvgarg-creator]

What does this PR do?

This PR objective is same as #13179
Fixes #13148
The issue was resolved for every tokenizer except the LayoutLmV2 tokenizer.

Before submitting

• Did you read the contributor guideline,
  Pull Request section? Yes 👍🏻.
• Was this discussed/approved via a Github issue or the forum? Yes👍🏻 , Slow tokenizers return overflowing tokens in reversed order #13148
• Did you make sure to update the documentation with your changes? Yes 👍🏻.
• Did you add any necessary tests?
  The following test have been added which check the sequence of overflowing tokens, bbox sequence and input_ids.
  • test_maximum_encoding_length_pair_input
  • test_maximum_encoding_length_single_input

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@LysandreJik @sgugger @SaulLu @NielsRogge

[SaulLu]

Thank you very much for the PR.

As it seems to me that we didn't see any tests failing for LayoutLMv2 at the time your previous PR was merged, could you confirm that this behaviour is tested for LayoutLMv2 now? As I see that in the PR you did not change any tests, I'd like to make sure we catch the problem with the tests next time 🙂 .

[Apoorvgarg-creator]

As I see that in the PR you did not change any tests, I'd like to make sure we catch the problem with the tests next time 🙂 .

@SaulLu, The test test_maximum_encoding_length_pair_input and test_maximum_encoding_length_single_input were skipped because of the reason LayoutLMv2 tokenizer requires boxes besides sequences.
I will make the necessary changes to the file test_tokenization_layoutlmv2.py to ensure that we catch the problem next time.
Thank you

[Apoorvgarg-creator]

@SaulLu, I have added a function get_clean_sequences which gives us list of words along with their ids and boxes, have taken reference from the test_tokenization_common.py.
Furthermore, I have modified the test_maximum_encoding_length_single_input to ensure the correct order of overflowing order.
Please can you review the changes.
Thank you

[SaulLu]

Thank you very much for all this work @Apoorvgarg-creator. 🎊

I left you mostly comments about the testing part. Especially as LayoutLmV2 has extra boxes, I think we should also add checks on the corresponding sequences.

Finally, if I'm not mistaken, there is no test_maximum_encoding_length_pair_input test for lajoutlmv2, it would be nice to have it too. 😄

Let me know if you need some help!

[Apoorvgarg-creator]

@SaulLu, I have gone through the reviews. I will make the changes.
I didn't commit test_maximum_encoding_length_pair_input function because I wanted to see the changes that are required in the get_clean_sequences and test_maximum_encoding_length_single_input.
Thank you for reviewing and suggesting the changes.

[Apoorvgarg-creator]

@SaulLu @NielsRogge @LysandreJik, Sorry for the delay in the PR. I am done with most of the work, but I would like to ask for help in Comparing bbox sequence for fast tokenizer for test_maximum_encoding_length_pair_input. That test isn't working properly. Can you help me from where I should read about this ?
I am sorry once again for delaying the work.
Update: I have resolved the problem I was facing.

[Apoorvgarg-creator]

@SaulLu, Working on test_maximum_encoding_length_pair_input. Will Commit the changes by end of the day.

[Apoorvgarg-creator]

@SaulLu @LysandreJik @NielsRogge, could you please review the test added and changes.
Thank you

[Apoorvgarg-creator]

@patrickvonplaten @SaulLu @LysandreJik @NielsRogge , Could you please review the PR.
Thank you.

[LysandreJik]

Thank you for the very extensive tests, great work!
Regarding the format:

• Could you add some comments explaining what is done? There is a lot of code so I fear it will become unmaintainable without clear comments explaining what is done.
• Please replace all assert with unittest's specific methods.

I'll let @NielsRogge review the logic.

[Apoorvgarg-creator]

• Could you add some comments explaining what is done? There is a lot of code so I fear it will become unmaintainable without clear comments explaining what is done.

Sure, Thank you for reviewing.

[NielsRogge]

This PR looks good to me.

Pinging @SaulLu for a final review.

[SaulLu]

Thank you very much for all your work @Apoorvgarg-creator !

Thanks especially for adding all those tests! I have one last little request which is about the overflowing boxes check, but to save you time I have opened a PR on your branch which includes the changes I am referring to. If you like these proposals, don't hesitate to merge this PR in your branch. 🙂

[SaulLu]

Thank you for these latest changes! It's all good for me!

[Apoorvgarg-creator]

@SaulLu @LysandreJik, I have been getting these mails, I tried checking the reason for their failing but couldn't understand the issue.
Message : This request was automatically failed because there were no enabled runners online to process the request for more than 1 days.
Here the Screenshot

[sgugger]

Thnaks a gain for all your work on this!