Call _set_save_spec() when creating TF models (#19321)

* Add a build_from_serving_sig_and_dummies method and replace all calls like model(model.dummy_inputs) with it.

* make fixup

* Remove the overridden save() as this is no longer necessary

* Also call _set_save_spec(), the last missing piece

* Ensure we set the save spec when loading from config too

* Turn this whole thing into a one-line PR

* Turn this whole thing into a one-line PR

* Turn this whole thing into a one-line PR
Co-authored-by: Your Name <you@example.com>

src/transformers/modeling_tf_utils.py

@@ -1049,6 +1049,8 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
         # Save config and origin of the pretrained weights if given in model
         self.config = config
         self.name_or_path = config.name_or_path
+        # Set the serving spec quickly to ensure that Keras doesn't use the specific dummy input shapes as the spec
+        self._set_save_spec(self.serving.input_signature[0])
 
     def get_config(self):
         return self.config.to_dict()
@@ -1097,29 +1099,6 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
         """
         raise NotImplementedError
 
-    def save(
-        self,
-        filepath,
-        overwrite=True,
-        include_optimizer=True,
-        save_format=None,
-        signatures=None,
-        options=None,
-        save_traces=True,
-    ):
-        # Very simple wrapper that ensures we set the correct serving signature when saving
-        if signatures is None and hasattr(self, "serving"):
-            signatures = self.serving
-        super().save(
-            filepath,
-            overwrite=overwrite,
-            include_optimizer=include_optimizer,
-            save_format=save_format,
-            signatures=signatures,
-            options=options,
-            save_traces=save_traces,
-        )
-
     def get_input_embeddings(self) -> tf.keras.layers.Layer:
         """
         Returns the model's input embeddings layer.