Stop storing references to bound methods via tf.function (#24146)

* Stop storing references to bound methods in tf.functions

* Remove the gc.collect calls now that we resolved the underlying problem

* Remove the default signature from model.serving entirely, big cleanup

* Remove _prune_signature as self.input_signature can prune itself

* Restore serving docstring

* Update int support test to check the input signature

* Make sure other tests also use model.input_signature and not serving.input_signature

* Restore _prune_signature

* Remove the doctest GC now it's no longer needed

* Correct core tests to use the pruned sig

* order lines correctly in core tests

* Add eager_serving back with a deprecation warning

tests/test_modeling_tf_common.py

@@ -1687,14 +1687,10 @@ class TFModelTesterMixin:
                 if tensor.dtype.is_integer:
                     self.assertTrue(tensor.dtype == tf.int32, "Integer dummy inputs should be tf.int32!")
 
-            # Also confirm that the serving sig uses int32
-            if hasattr(model, "serving"):
-                serving_sig = model.serving.input_signature
-                for key, tensor_spec in serving_sig[0].items():
-                    if tensor_spec.dtype.is_integer:
-                        self.assertTrue(
-                            tensor_spec.dtype == tf.int32, "Serving signatures should use tf.int32 for ints!"
-                        )
+            # Also confirm that the input_signature uses int32
+            for key, tensor_spec in model.input_signature.items():
+                if tensor_spec.dtype.is_integer:
+                    self.assertTrue(tensor_spec.dtype == tf.int32, "Input signatures should use tf.int32 for ints!")
 
     def test_generate_with_headmasking(self):
         attention_names = ["encoder_attentions", "decoder_attentions", "cross_attentions"]