Fix seq2seq doc tests (#16606)

* fix bart and mbart

* add ckpt names as variables

* fix mbart

* fix plbart

* use varibale for ckot name

src/transformers/models/bart/modeling_bart.py

@@ -56,11 +56,14 @@ _TOKENIZER_FOR_DOC = "BartTokenizer"
 _EXPECTED_OUTPUT_SHAPE = [1, 8, 768]
 
 # SequenceClassification docstring
-_SEQ_CLASS_EXPECTED_OUTPUT_SHAPE = [1, 2]
+_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATION = "valhalla/bart-large-sst2"
+_SEQ_CLASS_EXPECTED_LOSS = 0.0
+_SEQ_CLASS_EXPECTED_OUTPUT = "'POSITIVE'"
 
 # QuestionAsnwering docstring
-_QA_EXPECTED_LOSS = 2.98
-_QA_EXPECTED_OUTPUT_SHAPE = [1, 17]
+_CHECKPOINT_FOR_QA = "valhalla/bart-large-finetuned-squadv1"
+_QA_EXPECTED_LOSS = 0.59
+_QA_EXPECTED_OUTPUT = "' nice puppet'"
 
 
 BART_PRETRAINED_MODEL_ARCHIVE_LIST = [
@@ -1447,10 +1450,11 @@ class BartForSequenceClassification(BartPretrainedModel):
     @add_start_docstrings_to_model_forward(BART_INPUTS_DOCSTRING)
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
-        checkpoint=_CHECKPOINT_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATION,
         output_type=Seq2SeqSequenceClassifierOutput,
         config_class=_CONFIG_FOR_DOC,
-        expected_output=_SEQ_CLASS_EXPECTED_OUTPUT_SHAPE,
+        expected_output=_SEQ_CLASS_EXPECTED_OUTPUT,
+        expected_loss=_SEQ_CLASS_EXPECTED_LOSS,
     )
     def forward(
         self,
@@ -1572,11 +1576,11 @@ class BartForQuestionAnswering(BartPretrainedModel):
     @add_start_docstrings_to_model_forward(BART_INPUTS_DOCSTRING)
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
-        checkpoint=_CHECKPOINT_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_QA,
         output_type=Seq2SeqQuestionAnsweringModelOutput,
         config_class=_CONFIG_FOR_DOC,
         expected_loss=_QA_EXPECTED_LOSS,
-        expected_output=_QA_EXPECTED_OUTPUT_SHAPE,
+        expected_output=_QA_EXPECTED_OUTPUT,
     )
     def forward(
         self,

src/transformers/models/bigbird_pegasus/modeling_bigbird_pegasus.py

@@ -51,17 +51,20 @@ logger = logging.get_logger(__name__)
 
 _CHECKPOINT_FOR_DOC = "google/bigbird-pegasus-large-arxiv"
 _CONFIG_FOR_DOC = "BigBirdPegasusConfig"
-_TOKENIZER_FOR_DOC = "PegasusTokenizer"
+_TOKENIZER_FOR_DOC = "PegasusTokenizerFast"
 
 # Base model docstring
 _EXPECTED_OUTPUT_SHAPE = [1, 7, 1024]
 
 # SequenceClassification docstring
-_SEQ_CLASS_EXPECTED_OUTPUT_SHAPE = [1, 2]
+_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATION = "hf-internal-testing/tiny-random-bigbird_pegasus"
+_SEQ_CLASS_EXPECTED_LOSS = 0.69
+_SEQ_CLASS_EXPECTED_OUTPUT = "'LABEL_1'"
 
 # QuestionAsnwering docstring
-_QA_EXPECTED_LOSS = 2.56
-_QA_EXPECTED_OUTPUT_SHAPE = [1, 12]
+_CHECKPOINT_FOR_QA = "hf-internal-testing/tiny-random-bigbird_pegasus"
+_QA_EXPECTED_LOSS = 3.96
+_QA_EXPECTED_OUTPUT = "''"
 
 
 BIGBIRD_PEGASUS_PRETRAINED_MODEL_ARCHIVE_LIST = [
@@ -2645,10 +2648,11 @@ class BigBirdPegasusForSequenceClassification(BigBirdPegasusPreTrainedModel):
     @add_start_docstrings_to_model_forward(BIGBIRD_PEGASUS_INPUTS_DOCSTRING)
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
-        checkpoint=_CHECKPOINT_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATION,
         output_type=Seq2SeqSequenceClassifierOutput,
         config_class=_CONFIG_FOR_DOC,
-        expected_output=_SEQ_CLASS_EXPECTED_OUTPUT_SHAPE,
+        expected_output=_SEQ_CLASS_EXPECTED_OUTPUT,
+        expected_loss=_SEQ_CLASS_EXPECTED_LOSS,
     )
     def forward(
         self,
@@ -2771,11 +2775,11 @@ class BigBirdPegasusForQuestionAnswering(BigBirdPegasusPreTrainedModel):
     @add_start_docstrings_to_model_forward(BIGBIRD_PEGASUS_INPUTS_DOCSTRING)
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
-        checkpoint=_CHECKPOINT_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_QA,
         output_type=Seq2SeqQuestionAnsweringModelOutput,
         config_class=_CONFIG_FOR_DOC,
         expected_loss=_QA_EXPECTED_LOSS,
-        expected_output=_QA_EXPECTED_OUTPUT_SHAPE,
+        expected_output=_QA_EXPECTED_OUTPUT,
     )
     def forward(
         self,

src/transformers/models/mbart/modeling_mbart.py

@@ -55,11 +55,14 @@ _TOKENIZER_FOR_DOC = "MBartTokenizer"
 _EXPECTED_OUTPUT_SHAPE = [1, 8, 1024]
 
 # SequenceClassification docstring
-_SEQ_CLASS_EXPECTED_OUTPUT_SHAPE = [1, 2]
+_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATION = "hf-internal-testing/tiny-random-mbart"
+_SEQ_CLASS_EXPECTED_LOSS = 0.69
+_SEQ_CLASS_EXPECTED_OUTPUT = "'LABEL_1'"
 
 # QuestionAsnwering docstring
-_QA_EXPECTED_LOSS = 3.04
-_QA_EXPECTED_OUTPUT_SHAPE = [1, 16]
+_CHECKPOINT_FOR_QA = "hf-internal-testing/tiny-random-mbart"
+_QA_EXPECTED_LOSS = 3.55
+_QA_EXPECTED_OUTPUT = "'? Jim Henson was a'"
 
 
 MBART_PRETRAINED_MODEL_ARCHIVE_LIST = [
@@ -1437,10 +1440,11 @@ class MBartForSequenceClassification(MBartPreTrainedModel):
     @add_start_docstrings_to_model_forward(MBART_INPUTS_DOCSTRING)
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
-        checkpoint=_CHECKPOINT_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATION,
         output_type=Seq2SeqSequenceClassifierOutput,
         config_class=_CONFIG_FOR_DOC,
-        expected_output=_SEQ_CLASS_EXPECTED_OUTPUT_SHAPE,
+        expected_output=_SEQ_CLASS_EXPECTED_OUTPUT,
+        expected_loss=_SEQ_CLASS_EXPECTED_LOSS,
     )
     # Copied from transformers.models.bart.modeling_bart.BartForSequenceClassification.forward
     def forward(
@@ -1563,11 +1567,11 @@ class MBartForQuestionAnswering(MBartPreTrainedModel):
     @add_start_docstrings_to_model_forward(MBART_INPUTS_DOCSTRING)
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
-        checkpoint=_CHECKPOINT_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_QA,
         output_type=Seq2SeqQuestionAnsweringModelOutput,
         config_class=_CONFIG_FOR_DOC,
         expected_loss=_QA_EXPECTED_LOSS,
-        expected_output=_QA_EXPECTED_OUTPUT_SHAPE,
+        expected_output=_QA_EXPECTED_OUTPUT,
     )
     # Copied from transformers.models.bart.modeling_bart.BartForQuestionAnswering.forward
     def forward(

src/transformers/models/plbart/modeling_plbart.py

@@ -54,7 +54,9 @@ _TOKENIZER_FOR_DOC = "PLBartTokenizer"
 _EXPECTED_OUTPUT_SHAPE = [1, 8, 768]
 
 # SequenceClassification docstring
-_SEQ_CLASS_EXPECTED_OUTPUT_SHAPE = [1, 2]
+_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATION = "hf-internal-testing/tiny-plbart"
+_SEQ_CLASS_EXPECTED_OUTPUT = "'LABEL_1'"
+_SEQ_CLASS_EXPECTED_LOSS = 0.69
 
 
 PLBART_PRETRAINED_MODEL_ARCHIVE_LIST = [
@@ -1408,10 +1410,11 @@ class PLBartForSequenceClassification(PLBartPreTrainedModel):
     @add_start_docstrings_to_model_forward(PLBART_INPUTS_DOCSTRING)
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
-        checkpoint=_CHECKPOINT_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATION,
         output_type=Seq2SeqSequenceClassifierOutput,
         config_class=_CONFIG_FOR_DOC,
-        expected_output=_SEQ_CLASS_EXPECTED_OUTPUT_SHAPE,
+        expected_output=_SEQ_CLASS_EXPECTED_OUTPUT,
+        expected_loss=_SEQ_CLASS_EXPECTED_LOSS,
     )
     # Copied from transformers.models.bart.modeling_bart.BartForSequenceClassification.forward
     def forward(