Fix saved classifier models from before 0.14 (#1839)

We switched the class name for `XXClassifier` models to
`XXTextClasssifier`. However, a saved classifier before 0.16 will still
be looking for the class under the old name.

This updates our export helper to also registered the old name, so we can
restore to the new class when loading the model.

I also try to improve our error messages when we do encounter an
unrecognized class.

keras_nlp/src/api_export.py

@@ -22,7 +22,16 @@
     namex = None
 
 
-def maybe_register_serializable(symbol):
+def maybe_register_serializable(path, symbol):
+    # If we have multiple export names, actually make sure to register these
+    # first. This makes sure we have a backward compat mapping of old serialized
+    # name to new class.
+    if isinstance(path, (list, tuple)):
+        for name in path:
+            name = name.split(".")[-1]
+            keras.saving.register_keras_serializable(
+                package="keras_nlp", name=name
+            )(symbol)
     if isinstance(symbol, types.FunctionType) or hasattr(symbol, "get_config"):
         # We register twice, first with the old name, second with the new name,
         # so loading still works under the old name.
@@ -39,7 +48,7 @@ def __init__(self, path):
             super().__init__(package="keras_nlp", path=path)
 
         def __call__(self, symbol):
-            maybe_register_serializable(symbol)
+            maybe_register_serializable(self.path, symbol)
             return super().__call__(symbol)
 
 else:

keras_nlp/src/models/bert/bert_text_classifier_test.py

@@ -67,6 +67,15 @@ def test_saved_model(self):
             input_data=self.input_data,
         )
 
+    @pytest.mark.large
+    def test_smallest_preset(self):
+        self.run_preset_test(
+            cls=BertTextClassifier,
+            preset="bert_tiny_en_uncased_sst2",
+            input_data=self.input_data,
+            expected_output_shape=(2, 2),
+        )
+
     @pytest.mark.extra_large
     def test_all_presets(self):
         for preset in BertTextClassifier.presets:

keras_nlp/src/utils/preset_utils.py

@@ -569,7 +569,16 @@ def load_serialized_object(config, **kwargs):
 
 def check_config_class(config):
     """Validate a preset is being loaded on the correct class."""
-    return keras.saving.get_registered_object(config["registered_name"])
+    registered_name = config["registered_name"]
+    cls = keras.saving.get_registered_object(registered_name)
+    if cls is None:
+        raise ValueError(
+            f"Attempting to load class {registered_name} with "
+            "`from_preset()`, but there is no class registered with Keras "
+            f"for {registered_name}. Make sure to register any custom "
+            "classes with `register_keras_serializable()`."
+        )
+    return cls
 
 
 def jax_memory_cleanup(layer):

keras_nlp/src/utils/preset_utils_test.py

@@ -12,8 +12,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import json
 import os
 
+import keras
 import pytest
 from absl.testing import parameterized
 
@@ -38,6 +40,15 @@ def test_preset_errors(self):
         with self.assertRaisesRegex(ValueError, "Unknown preset identifier"):
             AlbertTextClassifier.from_preset("snaggle://bort/bort/bort")
 
+        backbone = BertBackbone.from_preset("bert_tiny_en_uncased")
+        preset_dir = self.get_temp_dir()
+        config = keras.utils.serialize_keras_object(backbone)
+        config["registered_name"] = "keras_nlp>BortBackbone"
+        with open(os.path.join(preset_dir, CONFIG_FILE), "w") as config_file:
+            config_file.write(json.dumps(config, indent=4))
+        with self.assertRaisesRegex(ValueError, "class keras_nlp>BortBackbone"):
+            BertBackbone.from_preset(preset_dir)
+
     def test_upload_empty_preset(self):
         temp_dir = self.get_temp_dir()
         empty_preset = os.path.join(temp_dir, "empty")