[Consistency] Automatically set decoder_input_ids for TFEncoderDecoderModel

[NielsRogge]

🚀 Feature request

Currently, the EncoderDecoderModel class in PyTorch automatically creates the decoder_input_ids based on the labels provided by the user (similar to how this is done for T5/BART). This should also be implemented for TFEncoderDecoderModel, because currently users should manually provide decoder_input_ids to the model.

One can take a look at the TF implementation of BART for example to see how to shift the labels in order to automatically create the decoder_input_ids, namely here.

One should then also update the docstring correspondingly.

[Carlosbogo]

Hi, I would like to work on this. Is there anyone already working on it?

[NielsRogge]

No, you can work on it if you want!

[ankushbhatia2]

Hi, Is it still open? I can work on this issue

[NielsRogge]

There's a PR open (#14469), however it's not finished yet.

[katiele47]

Hi @NielsRogge Can I finish up this issue with the closed PR as reference? I'd like to make a PR soon.

[hackyon]

I believe this has already been resolved in #15175.

[byi8220]

It appears that way. I am new to this codebase (and looking for a good first contribution), so take any of my opinion with a grain of salt, but this block appears to implement this functionality as part of the PR:

transformers/src/transformers/models/vision_encoder_decoder/modeling_tf_vision_encoder_decoder.py

Lines 572 to 575 in 831bc25

	if (labels is not None) and (decoder_input_ids is None and decoder_inputs_embeds is None):
	decoder_input_ids = shift_tokens_right(
	labels, self.config.pad_token_id, self.config.decoder_start_token_id
	)

If this is true, would it make sense to close this issue?

[NielsRogge]

Closing this one as it has ben resolved by #15175. Thanks for reporting