Fix OPT doc tests

[ArthurZucker]

What does this PR do?

Fixes the doctest for OPT by adding slight modifications.

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[ydshieh]

Hi @ArthurZucker Do you know why the expected value has extra ``` ? OPTForCausalLM has the same expected value, but it doesn't have test failure. Seems strange to me 😕

 1003         >>> # Generate
 1004         >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
 1005         >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
 Expected:
     "Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."
     ```
 Got:
     "Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."

[ArthurZucker]

Actually it seems that the automatic doc préparation that adds a line does not affect the TFOPTForCausalLM forward doc. Thus I removed the manually created example and defeated to using add_code_sample which worked in that case.
Now why does the add extra line does not work is another mystery and I could not setup a debugging instance to check that

[ydshieh]

Thank you. I will check this on Monday.

[ydshieh]

Thank you for the fix, @ArthurZucker ! LGTM, especially all tests pass now.
Let's wait an approval from some core maintainers.

[ydshieh]

Regarding the mystery , it is due to this line

``decoder_input_ids```

transformers/src/transformers/models/opt/modeling_tf_opt.py

Line 601 in b2e4b09

all ``decoder_input_ids``` of shape `(batch_size, sequence_length)`.

We can fix this in this PR too :-) It should be simply

`decoder_input_ids`

[ArthurZucker]

I forgot to change the name in the add_docstring will fix that

[sgugger]

Thanks for fixing!