Fix EncoderDecoderModel docs (#14197)

* Fix docs

* Apply suggestions from review + fix bug

src/transformers/models/encoder_decoder/modeling_encoder_decoder.py

@@ -98,9 +98,9 @@ ENCODER_DECODER_INPUTS_DOCSTRING = r"""
             If :obj:`past_key_values` is used, optionally only the last :obj:`decoder_input_ids` have to be input (see
             :obj:`past_key_values`).
 
-            Provide for sequence to sequence training to the decoder. Indices can be obtained using
-            :class:`~transformers.PreTrainedTokenizer`. See :meth:`transformers.PreTrainedTokenizer.encode` and
-            :meth:`transformers.PreTrainedTokenizer.__call__` for details.
+            For training, :obj:`decoder_input_ids` are automatically created by the model by shifting the :obj:`labels`
+            to the right, replacing -100 by the :obj:`pad_token_id` and prepending them with the
+            :obj:`decoder_start_token_id`.
         decoder_attention_mask (:obj:`torch.BoolTensor` of shape :obj:`(batch_size, target_sequence_length)`, `optional`):
             Default behavior: generate a tensor that ignores pad tokens in :obj:`decoder_input_ids`. Causal mask will
             also be used by default.
@@ -425,12 +425,14 @@ class EncoderDecoderModel(PreTrainedModel):
             >>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
             >>> model = EncoderDecoderModel.from_encoder_decoder_pretrained('bert-base-uncased', 'bert-base-uncased') # initialize Bert2Bert from pre-trained checkpoints
 
-            >>> # forward
-            >>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
-            >>> outputs = model(input_ids=input_ids, decoder_input_ids=input_ids)
-
             >>> # training
-            >>> outputs = model(input_ids=input_ids, decoder_input_ids=input_ids, labels=input_ids)
+            >>> model.config.decoder_start_token_id = tokenizer.cls_token_id
+            >>> model.config.pad_token_id = tokenizer.pad_token_id
+            >>> model.config.vocab_size = model.config.decoder.vocab_size
+
+            >>> input_ids = tokenizer("Hello, my dog is cute", return_tensors="pt").input_ids
+            >>> labels = tokenizer("Salut, mon chien est mignon", return_tensors="pt").input_ids
+            >>> outputs = model(input_ids=input_ids, labels=input_ids)
             >>> loss, logits = outputs.loss, outputs.logits
 
             >>> # save and load from pretrained

src/transformers/models/speech_encoder_decoder/modeling_speech_encoder_decoder.py

@@ -103,9 +103,9 @@ SPEECH_ENCODER_DECODER_INPUTS_DOCSTRING = r"""
             If :obj:`past_key_values` is used, optionally only the last :obj:`decoder_input_ids` have to be input (see
             :obj:`past_key_values`).
 
-            Provide for sequence to sequence training to the decoder. Indices can be obtained using
-            :class:`~transformers.PreTrainedTokenizer`. See :meth:`transformers.PreTrainedTokenizer.encode` and
-            :meth:`transformers.PreTrainedTokenizer.__call__` for details.
+            For training, :obj:`decoder_input_ids` are automatically created by the model by shifting the :obj:`labels`
+            to the right, replacing -100 by the :obj:`pad_token_id` and prepending them with the
+            :obj:`decoder_start_token_id`.
         decoder_attention_mask (:obj:`torch.BoolTensor` of shape :obj:`(batch_size, target_sequence_length)`, `optional`):
             Default behavior: generate a tensor that ignores pad tokens in :obj:`decoder_input_ids`. Causal mask will
             also be used by default.
@@ -424,25 +424,19 @@ class SpeechEncoderDecoderModel(PreTrainedModel):
         Examples::
 
             >>> from transformers import SpeechEncoderDecoderModel, Speech2Text2Processor
+            >>> from datasets import load_dataset
             >>> import torch
 
             >>> processor = Speech2Text2Processor.from_pretrained('facebook/s2t-wav2vec2-large-en-de')
             >>> model = SpeechEncoderDecoderModel.from_pretrained('facebook/s2t-wav2vec2-large-en-de')
 
-            >>> # process dataset
-            >>> def map_to_array(batch):
-            >>>     speech, _ = sf.read(batch["file"])
-            >>>     batch["speech"] = speech
-            >>>     return batch
-
             >>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
-            >>> ds = ds.map(map_to_array)
 
-            >>> input_values = processor(ds["speech"][0], return_tensors="pt").input_values  # Batch size 1
+            >>> input_values = processor(ds[0]["audio"]["array"], return_tensors="pt").input_values
             >>> decoder_input_ids = torch.tensor([[model.config.decoder.decoder_start_token_id]])
             >>> outputs = model(input_values=input_values, decoder_input_ids=decoder_input_ids)
 
-            >>> # generation
+            >>> # inference (generation)
             >>> generated = model.generate(input_values)
             >>> translation = processor.batch_decode(generated)
 

src/transformers/models/vision_encoder_decoder/modeling_vision_encoder_decoder.py

@@ -113,9 +113,9 @@ VISION_ENCODER_DECODER_INPUTS_DOCSTRING = r"""
             If :obj:`past_key_values` is used, optionally only the last :obj:`decoder_input_ids` have to be input (see
             :obj:`past_key_values`).
 
-            Provide for sequence to sequence training to the decoder. Indices can be obtained using
-            :class:`~transformers.PreTrainedTokenizer`. See :meth:`transformers.PreTrainedTokenizer.encode` and
-            :meth:`transformers.PreTrainedTokenizer.__call__` for details.
+            For training, :obj:`decoder_input_ids` are automatically created by the model by shifting the :obj:`labels`
+            to the right, replacing -100 by the :obj:`pad_token_id` and prepending them with the
+            :obj:`decoder_start_token_id`.
         decoder_attention_mask (:obj:`torch.BoolTensor` of shape :obj:`(batch_size, target_sequence_length)`, `optional`):
             Default behavior: generate a tensor that ignores pad tokens in :obj:`decoder_input_ids`. Causal mask will
             also be used by default.
@@ -428,9 +428,15 @@ class VisionEncoderDecoderModel(PreTrainedModel):
             >>> image = Image.open(requests.get(url, stream=True).raw).convert("RGB")
 
             >>> # training
-            >>> pixel_values = processor(image, return_tensors="pt").pixel_values  # Batch size 1
-            >>> decoder_input_ids = torch.tensor([[model.config.decoder.decoder_start_token_id]])
-            >>> outputs = model(pixel_values=pixel_values, decoder_input_ids=decoder_input_ids)
+            >>> model.config.decoder_start_token_id = processor.tokenizer.cls_token_id
+            >>> model.config.pad_token_id = processor.tokenizer.pad_token_id
+            >>> model.config.vocab_size = model.config.decoder.vocab_size
+
+            >>> pixel_values = processor(image, return_tensors="pt").pixel_values
+            >>> text = "hello world"
+            >>> labels = processor.tokenizer(text, return_tensors="pt").input_ids
+            >>> outputs = model(pixel_values=pixel_values, labels=labels)
+            >>> loss = outputs.loss
 
             >>> # inference (generation)
             >>> generated_ids = model.generate(pixel_values)