Does MBartTokenizer remove the parameter decoder_input_ids?

[wmathor]

Environment info

• transformers version:3.4.0
• Platform:Google Colab
• Python version:3.7
• PyTorch version (GPU?):1.7.0+cu101
• Tensorflow version (GPU?):2.x
• Using GPU in script?: no
• Using distributed or parallel set-up in script?:no

Who can help

Information

Model I am using (Bert, XLNet ...): mbart

The problem arises when using:

• the official example scripts: (give details below)
• my own modified scripts: (give details below)

The tasks I am working on is:

• an official GLUE/SQUaD task: (give the name)
• my own task or dataset: (give details below)

To reproduce

example_english_phrase = "UN Chief Says There Is No Military Solution in Syria"
expected_translation_romanian = "Şeful ONU declară că nu există o soluţie militară în Siria"
batch = MBartTokenizer.from_pretrained('facebook/mbart-large-cc25').prepare_seq2seq_batch(example_english_phrase, src_lang="en_XX", tgt_lang="ro_RO", tgt_texts=expected_translation_romanian)
input_ids = batch["input_ids"]
target_ids = batch["decoder_input_ids"]

Steps to reproduce the behavior:

KeyError                                  Traceback (most recent call last)
<ipython-input-11-b3eedaf10c3e> in <module>()
      3 batch = MBartTokenizer.from_pretrained('facebook/mbart-large-en-ro').prepare_seq2seq_batch(example_english_phrase, src_lang="en_XX", tgt_lang="ro_RO", tgt_texts=expected_translation_romanian)
      4 input_ids = batch["input_ids"]
----> 5 target_ids = batch["decoder_input_ids"]

/usr/local/lib/python3.6/dist-packages/transformers/tokenization_utils_base.py in __getitem__(self, item)
    232         """
    233         if isinstance(item, str):
--> 234             return self.data[item]
    235         elif self._encodings is not None:
    236             return self._encodings[item]

KeyError: 'decoder_input_ids'

[sshleifer]

The docs are incorrect, sorry about that.

Try

    from transformers import MBartForConditionalGeneration, MBartTokenizer
    model = MBartForConditionalGeneration.from_pretrained("facebook/mbart-large-en-ro")
    tokenizer = MBartTokenizer.from_pretrained("facebook/mbart-large-en-ro")
    article = "UN Chief Says There Is No Military Solution in Syria"
    batch = tokenizer.prepare_seq2seq_batch(src_texts=[article], src_lang="en_XX")
    translated_tokens = model.generate(**batch, decoder_start_token_id=tokenizer.lang_code_to_id["ro_RO"])
    translation = tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)[0]
    assert translation == "Şeful ONU declară că nu există o soluţie militară în Siria"

[wmathor]

The docs are incorrect, sorry about that.

Try

    from transformers import MBartForConditionalGeneration, MBartTokenizer
    model = MBartForConditionalGeneration.from_pretrained("facebook/mbart-large-en-ro")
    tokenizer = MBartTokenizer.from_pretrained("facebook/mbart-large-en-ro")
    article = "UN Chief Says There Is No Military Solution in Syria"
    batch = tokenizer.prepare_seq2seq_batch(src_texts=[article], src_lang="en_XX")
    translated_tokens = model.generate(**batch, decoder_start_token_id=tokenizer.lang_code_to_id["ro_RO"])
    translation = tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)[0]
    assert translation == "Şeful ONU declară că nu există o soluţie militară în Siria"

thank you for your reply, If I don't want to generate, I just want to train. How should I change it?

example_english_phrase = "UN Chief Says There Is No Military Solution in Syria"
expected_translation_romanian = "Şeful ONU declară că nu există o soluţie militară în Siria"
batch = tokenizer.prepare_seq2seq_batch(example_english_phrase, src_lang="en_XX", tgt_lang="ro_RO", tgt_texts=expected_translation_romanian)
input_ids = batch["input_ids"]
target_ids = batch["decoder_input_ids"] # Error
decoder_input_ids = target_ids[:, :-1].contiguous()
labels = target_ids[:, 1:].clone()
model(input_ids=input_ids, decoder_input_ids=decoder_input_ids, labels=labels) #forward

[sshleifer]

See this https://github.com/huggingface/transformers/blob/master/examples/seq2seq/finetune.py#L138

the batch argument to that fn is the same as your batch (the output of prepare_seq2seq_batch)