add more examples[support dynamic, static and aware_training quantization]

[xin3he]

What does this PR do?

This PR will provide more available tasks for user to apply. Tasks can be apply with dynamic, static or aware_training quantization with INC.

Task includes:

• [multiple-choice] Done
• [summarization] Done
• [language-modeling] CLM Done, MLM&PLM Done.
  • limitation: a seed set in MLM&PLM take_eval_step to ensure verify loading success.
• [translation] Done.
  issues#14268, PR#14276 fixed the issue.

[xin3he]

Is there a test for examples to trigger?
What should I do to make it happen?
Could you provide a template for readme.MD？
BTW, [language-modeling] and [translation] are not ready yet. Please wait, thanks.

[xin3he]

According to our local test, we should upgrade the minimum version of transformers to 4.10.0 to avoid some errors in fx mode.
FX submodule naming fix

[xin3he]

UT failed because PyTorch1.10 is released now. Transformers requires PyTorch 1.9.

[echarlaix]

Thanks for the new examples @xin3he, I am currently looking at adding other examples on my side as well (question-answering, token-classification). For the README.md, we are going to follow something close to what was done in transformers such as done for question-answering or summarization. Finally concerning tests for examples, it was not planned initially but we can definitely add it, would you like to discuss it in our next meeting ? If you prefer to have an example before that I can of course create one.

[xin3he]

Hi, @echarlaix, an example of tests will be better for our discussion, thank you. We can discuss it in our next meeting.

[changwangss]

MLM/PLM verifiy_loading failed due to different accuracy between quantization and reloading, which is caused by plm/mlm_probablity leading to different input in reloading.