Remove the last few TF serving sigs (#23738)

Remove some more serving methods that (I think?) turned up while this PR was open

src/transformers/models/sam/modeling_tf_sam.py

@@ -1147,12 +1147,8 @@ class TFSamPreTrainedModel(TFPreTrainedModel):
 
     @property
     def dummy_inputs(self) -> Dict[str, tf.Tensor]:
-        """
+        # We override the default dummy inputs here because SAM has some really explosive memory usage in the
-        Dummy inputs to build the network.
+        # attention layers, so we want to pass the smallest possible batches
-
-        Returns:
-            `Dict[str, tf.Tensor]`: The dummy inputs.
-        """
         VISION_DUMMY_INPUTS = tf.random.uniform(
             shape=(
                 1,
@@ -1164,25 +1160,6 @@ class TFSamPreTrainedModel(TFPreTrainedModel):
         )
         return {"pixel_values": tf.constant(VISION_DUMMY_INPUTS)}
 
-    @tf.function(
-        input_signature=[
-            {
-                "pixel_values": tf.TensorSpec((None, None, None, None), tf.float32, name="pixel_values"),
-            }
-        ]
-    )
-    def serving(self, inputs):
-        """
-        Method used for serving the model.
-
-        Args:
-            inputs (`Dict[str, tf.Tensor]`):
-                The input of the saved model as a dictionary of tensors.
-        """
-        output = self.call(inputs)
-
-        return self.serving_output(output)
-
 
 SAM_START_DOCSTRING = r"""
     This model inherits from [`TFPreTrainedModel`]. Check the superclass documentation for the generic methods the

templates/adding_a_new_model/cookiecutter-template-{{cookiecutter.modelname}}/modeling_tf_{{cookiecutter.lowercase_modelname}}.py

@@ -1938,34 +1938,6 @@ class TF{{cookiecutter.camelcase_modelname}}PreTrainedModel(TFPreTrainedModel):
     config_class = {{cookiecutter.camelcase_modelname}}Config
     base_model_prefix = "model"
 
-    @property
-    def dummy_inputs(self):
-        pad_token = 1
-        input_ids = tf.cast(tf.convert_to_tensor(DUMMY_INPUTS), tf.int32)
-        decoder_input_ids = tf.cast(tf.convert_to_tensor(DUMMY_INPUTS), tf.int32)
-        dummy_inputs = {
-            "decoder_input_ids": decoder_input_ids,
-            "attention_mask": tf.math.not_equal(input_ids, pad_token),
-            "input_ids": input_ids,
-        }
-        return dummy_inputs
-
-    @tf.function(
-        input_signature=[
-            {
-                "input_ids": tf.TensorSpec((None, None), tf.int32, name="input_ids"),
-                "attention_mask": tf.TensorSpec((None, None), tf.int32, name="attention_mask"),
-                "decoder_input_ids": tf.TensorSpec((None, None), tf.int32, name="decoder_input_ids"),
-                "decoder_attention_mask": tf.TensorSpec((None, None), tf.int32, name="decoder_attention_mask"),
-            }
-        ]
-    )
-    # Copied from transformers.models.bart.modeling_tf_bart.TFBartPretrainedModel.serving
-    def serving(self, inputs):
-        output = self.call(inputs)
-
-        return self.serving_output(output)
-
 
 {{cookiecutter.uppercase_modelname}}_START_DOCSTRING = r"""
     This model inherits from [`TFPreTrainedModel`]. Check the superclass documentation for the