Can't use decoder_inputs_embeds argument on MBartForConditionalGeneration

[hibiki12y]

System Info

• transformers version: 4.22.0
• Platform: Linux-4.4.0-210-generic-x86_64-with-glibc2.23
• Python version: 3.9.12
• Huggingface_hub version: 0.7.0
• PyTorch version (GPU?): 1.12.0 (True)
• Tensorflow version (GPU?): 2.9.1 (True)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Using GPU in script?: yes
• Using distributed or parallel set-up in script?: yes

Who can help?

@patil-suraj

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

maybe not necessary

Expected behavior

#13800

Looks same problem at MBartForConditionalGeneration.foward

It is written as:

if decoder_input_ids is None:
    decoder_input_ids = shift_tokens_right(labels, self.config.pad_token_id)

So, looks need edit to:

if decoder_input_ids is None and decoder_inputs_embeds is None:
    decoder_input_ids = shift_tokens_right(labels, self.config.pad_token_id)

[LysandreJik]

Maybe of interest to @ArthurZucker :)

[ArthurZucker]

Having a look right now! Thanks for finding this 🤗