Fix OPT doc tests (#18365)

src/transformers/models/opt/modeling_opt.py

@@ -43,6 +43,11 @@ _TOKENIZER_FOR_DOC = "GPT2Tokenizer"
 # Base model docstring
 _EXPECTED_OUTPUT_SHAPE = [1, 8, 1024]
 
+# SequenceClassification docstring
+_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATION = "ArthurZ/opt-350m-dummy-sc"
+_SEQ_CLASS_EXPECTED_LOSS = 1.71
+_SEQ_CLASS_EXPECTED_OUTPUT = "'LABEL_0'"
+
 
 OPT_PRETRAINED_MODEL_ARCHIVE_LIST = [
     "facebook/opt-125m",
@@ -474,7 +479,6 @@ class OPTDecoder(OPTPreTrainedModel):
 
     Args:
         config: OPTConfig
-        embed_tokens (nn.Embedding): output embedding
     """
 
     def __init__(self, config: OPTConfig):
@@ -1008,10 +1012,11 @@ class OPTForSequenceClassification(OPTPreTrainedModel):
     @add_start_docstrings_to_model_forward(OPT_INPUTS_DOCSTRING)
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_SEQUENCE_CLASSIFICATION,
         output_type=SequenceClassifierOutputWithPast,
         config_class=_CONFIG_FOR_DOC,
-        expected_output="'LABEL_0'",
-        expected_loss=5.28,
+        expected_output=_SEQ_CLASS_EXPECTED_OUTPUT,
+        expected_loss=_SEQ_CLASS_EXPECTED_LOSS,
     )
     def forward(
         self,

src/transformers/models/opt/modeling_tf_opt.py

@@ -53,6 +53,9 @@ _TOKENIZER_FOR_DOC = "GPT2Tokenizer"
 # Base model docstring
 _EXPECTED_OUTPUT_SHAPE = [1, 8, 1024]
 
+# Causal LM output
+_CAUSAL_LM_EXPECTED_OUTPUT = "Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."
+
 LARGE_NEGATIVE = -1e8
 
 
@@ -894,6 +897,13 @@ class TFOPTForCausalLM(TFOPTPreTrainedModel, TFCausalLanguageModelingLoss):
 
     @unpack_inputs
     @replace_return_docstrings(output_type=TFCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=TFCausalLMOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+        expected_output=_CAUSAL_LM_EXPECTED_OUTPUT,
+    )
     def call(
         self,
         input_ids: Optional[TFModelInputType] = None,
@@ -964,25 +974,7 @@ class TFOPTForCausalLM(TFOPTPreTrainedModel, TFCausalLanguageModelingLoss):
                 for more detail.
             return_dict (`bool`, *optional*):
                 Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
-
-        Returns:
-
-        Example:
-
-        ```python
-        >>> from transformers import GPT2Tokenizer, TFOPTForCausalLM
-
-        >>> model = TFOPTForCausalLM.from_pretrained("facebook/opt-350m")
-        >>> tokenizer = GPT2Tokenizer.from_pretrained("facebook/opt-350m")
-
-        >>> prompt = "Hey, are you consciours? Can you talk to me?"
-        >>> inputs = tokenizer(prompt, return_tensors="tf")
-
-        >>> # Generate
-        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
-        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
-        "Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."
-        ```"""
+        """
 
         output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
         output_hidden_states = (