From 79fbddb108465bf6603349a0027d48e6d12e9d25 Mon Sep 17 00:00:00 2001
From: Matt Watson <mattdangerw@gmail.com>
Date: Tue, 3 Sep 2024 10:10:27 -0700
Subject: [PATCH 1/8] Add image and audio converter classes

These classes will occupy the same role as tokenizers for text models.
They will transform raw inputs to model inputs in a way that is not
task specific.
---
 keras_nlp/api/__init__.py                     |   1 -
 keras_nlp/api/layers/__init__.py              |   5 +
 keras_nlp/api/models/__init__.py              |   7 +-
 .../layers/preprocessing/audio_converter.py   | 108 ++++++
 .../layers/preprocessing/image_converter.py   | 150 +++++++++
 .../preprocessing/masked_lm_mask_generator.py |   4 +-
 .../preprocessing/multi_segment_packer.py     |   4 +-
 .../layers/preprocessing/random_deletion.py   |   4 +-
 .../src/layers/preprocessing/random_swap.py   |   4 +-
 .../layers/preprocessing/start_end_packer.py  |   4 +-
 .../src/models/bart/bart_preprocessor.py      |   4 +-
 .../src/models/bloom/bloom_preprocessor.py    |   4 +-
 .../src/models/causal_lm_preprocessor.py      |   8 +-
 .../deberta_v3_masked_lm_preprocessor.py      |   4 +-
 ...deberta_v3_text_classifier_preprocessor.py |   4 +-
 .../distil_bert_masked_lm_preprocessor.py     |   4 +-
 ...istil_bert_text_classifier_preprocessor.py |   4 +-
 .../models/electra/electra_preprocessor.py    |   4 +-
 .../f_net/f_net_masked_lm_preprocessor.py     |   4 +-
 .../f_net_text_classifier_preprocessor.py     |   4 +-
 .../src/models/falcon/falcon_preprocessor.py  |   4 +-
 .../src/models/gemma/gemma_preprocessor.py    |   4 +-
 .../src/models/gpt2/gpt2_preprocessor.py      |   4 +-
 .../gpt_neo_x/gpt_neo_x_preprocessor.py       |   4 +-
 .../src/models/llama/llama_preprocessor.py    |   4 +-
 .../src/models/masked_lm_preprocessor.py      |   4 +-
 .../models/mistral/mistral_preprocessor.py    |   4 +-
 keras_nlp/src/models/opt/opt_preprocessor.py  |   4 +-
 .../pali_gemma/pali_gemma_backbone_test.py    |  67 ++--
 .../pali_gemma_causal_lm_preprocessor.py      |  38 ++-
 .../pali_gemma_causal_lm_preprocessor_test.py |  15 +-
 .../pali_gemma/pali_gemma_causal_lm_test.py   |  13 +-
 .../pali_gemma/pali_gemma_image_converter.py} |  14 +-
 .../src/models/phi3/phi3_preprocessor.py      |   4 +-
 keras_nlp/src/models/preprocessor.py          |  50 ++-
 .../roberta/roberta_masked_lm_preprocessor.py |   4 +-
 .../roberta_text_classifier_preprocessor.py   |   4 +-
 .../src/models/seq_2_seq_lm_preprocessor.py   |   8 +-
 .../models/text_classifier_preprocessor.py    |   4 +-
 ...xtractor.py => whisper_audio_converter.py} |  31 +-
 ...est.py => whisper_audio_converter_test.py} |  10 +-
 .../models/whisper/whisper_preprocessor.py    | 315 ------------------
 .../whisper/whisper_preprocessor_test.py      |  85 -----
 .../xlm_roberta_masked_lm_preprocessor.py     |   4 +-
 ...lm_roberta_text_classifier_preprocessor.py |   4 +-
 .../src/tokenizers/byte_pair_tokenizer.py     |   6 +-
 keras_nlp/src/tokenizers/byte_tokenizer.py    |   6 +-
 .../tokenizers/sentence_piece_tokenizer.py    |   6 +-
 keras_nlp/src/tokenizers/tokenizer.py         |   4 +-
 .../tokenizers/unicode_codepoint_tokenizer.py |   6 +-
 .../src/tokenizers/word_piece_tokenizer.py    |   6 +-
 keras_nlp/src/utils/preset_utils.py           |  26 ++
 keras_nlp/src/utils/tensor_utils.py           |  55 ++-
 keras_nlp/src/utils/tensor_utils_test.py      |  29 +-
 54 files changed, 571 insertions(+), 610 deletions(-)
 create mode 100644 keras_nlp/src/layers/preprocessing/audio_converter.py
 create mode 100644 keras_nlp/src/layers/preprocessing/image_converter.py
 rename keras_nlp/{api/utils/__init__.py => src/models/pali_gemma/pali_gemma_image_converter.py} (61%)
 rename keras_nlp/src/models/whisper/{whisper_audio_feature_extractor.py => whisper_audio_converter.py} (92%)
 rename keras_nlp/src/models/whisper/{whisper_audio_feature_extractor_test.py => whisper_audio_converter_test.py} (84%)
 delete mode 100644 keras_nlp/src/models/whisper/whisper_preprocessor.py
 delete mode 100644 keras_nlp/src/models/whisper/whisper_preprocessor_test.py

diff --git a/keras_nlp/api/__init__.py b/keras_nlp/api/__init__.py
index 0225dd69a0..d0dc4576c6 100644
--- a/keras_nlp/api/__init__.py
+++ b/keras_nlp/api/__init__.py
@@ -22,7 +22,6 @@
 from keras_nlp.api import models
 from keras_nlp.api import samplers
 from keras_nlp.api import tokenizers
-from keras_nlp.api import utils
 from keras_nlp.src.utils.preset_utils import upload_preset
 from keras_nlp.src.version_utils import __version__
 from keras_nlp.src.version_utils import version
diff --git a/keras_nlp/api/layers/__init__.py b/keras_nlp/api/layers/__init__.py
index 73ad66b30b..5cffa3fb40 100644
--- a/keras_nlp/api/layers/__init__.py
+++ b/keras_nlp/api/layers/__init__.py
@@ -36,6 +36,8 @@
 )
 from keras_nlp.src.layers.modeling.transformer_decoder import TransformerDecoder
 from keras_nlp.src.layers.modeling.transformer_encoder import TransformerEncoder
+from keras_nlp.src.layers.preprocessing.audio_converter import AudioConverter
+from keras_nlp.src.layers.preprocessing.image_converter import ImageConverter
 from keras_nlp.src.layers.preprocessing.masked_lm_mask_generator import (
     MaskedLMMaskGenerator,
 )
@@ -45,3 +47,6 @@
 from keras_nlp.src.layers.preprocessing.random_deletion import RandomDeletion
 from keras_nlp.src.layers.preprocessing.random_swap import RandomSwap
 from keras_nlp.src.layers.preprocessing.start_end_packer import StartEndPacker
+from keras_nlp.src.models.pali_gemma.pali_gemma_image_converter import (
+    PaliGemmaImageConverter,
+)
diff --git a/keras_nlp/api/models/__init__.py b/keras_nlp/api/models/__init__.py
index e78271531c..7f752d1f02 100644
--- a/keras_nlp/api/models/__init__.py
+++ b/keras_nlp/api/models/__init__.py
@@ -228,13 +228,10 @@
 from keras_nlp.src.models.text_classifier_preprocessor import (
     TextClassifierPreprocessor,
 )
-from keras_nlp.src.models.whisper.whisper_audio_feature_extractor import (
-    WhisperAudioFeatureExtractor,
+from keras_nlp.src.models.whisper.whisper_audio_converter import (
+    WhisperAudioConverter,
 )
 from keras_nlp.src.models.whisper.whisper_backbone import WhisperBackbone
-from keras_nlp.src.models.whisper.whisper_preprocessor import (
-    WhisperPreprocessor,
-)
 from keras_nlp.src.models.whisper.whisper_tokenizer import WhisperTokenizer
 from keras_nlp.src.models.xlm_roberta.xlm_roberta_backbone import (
     XLMRobertaBackbone,
diff --git a/keras_nlp/src/layers/preprocessing/audio_converter.py b/keras_nlp/src/layers/preprocessing/audio_converter.py
new file mode 100644
index 0000000000..9569c44801
--- /dev/null
+++ b/keras_nlp/src/layers/preprocessing/audio_converter.py
@@ -0,0 +1,108 @@
+# Copyright 2024 The KerasNLP Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from keras_nlp.src.api_export import keras_nlp_export
+from keras_nlp.src.layers.preprocessing.preprocessing_layer import (
+    PreprocessingLayer,
+)
+from keras_nlp.src.utils.preset_utils import AUDIO_CONVERTER_CONFIG_FILE
+from keras_nlp.src.utils.preset_utils import find_subclass
+from keras_nlp.src.utils.preset_utils import get_preset_loader
+from keras_nlp.src.utils.preset_utils import save_serialized_object
+
+
+@keras_nlp_export("keras_nlp.layers.AudioConverter")
+class AudioConverter(PreprocessingLayer):
+    """Convert raw audio for models that support audio input.
+
+    This class converts from raw audio tensors of any length, to preprocessed
+    audio for pretrained model inputs. It is meant to be a convenient way to
+    write custom preprocessing code that is not model specific. This layer
+    should be instantiated via the `from_preset()` constructor, which will
+    create the correct subclass of this layer for the model preset.
+
+    The layer will take a input a raw audio tensor with shape `(batch_size,
+    num_samples)`, and output a preprocessed audio input for modeling. The exact
+    structure of the preprocessed input will vary per model. Preprocessing
+    will often include computing a spectogram of the raw audio signal.
+
+    Examples:
+    ```python
+    # Load an audio converter from a preset.
+    converter = keras_nlp.layers.AudioConverter.from_preset("whisper_base_en")
+    # Convert some raw audio input.
+    converter(np.ones(2, 1_000))
+    ```
+    """
+
+    @classmethod
+    def from_preset(
+        cls,
+        preset,
+        **kwargs,
+    ):
+        """Instantiate a `keras_nlp.layers.AudioConverter` from a model preset.
+
+        A preset is a directory of configs, weights and other file assets used
+        to save and load a pre-trained model. The `preset` can be passed as a
+        one of:
+
+        1. a built in preset identifier like `'whisper_base_en'`
+        2. a Kaggle Models handle like
+           `'kaggle://user/whisper/keras/whisper_base_en'`
+        3. a Hugging Face handle like `'hf://user/whisper_base_en'`
+        4. a path to a local preset directory like `'./whisper_base_en'`
+
+        You can run `cls.presets.keys()` to list all built-in presets available
+        on the class.
+
+        This constructor can be called in one of two ways. Either from the base
+        class like `keras_nlp.models.AudioConverter.from_preset()`, or from a
+        model class like `keras_nlp.models.WhisperAudioConverter.from_preset()`.
+        If calling from the base class, the subclass of the returning object
+        will be inferred from the config in the preset directory.
+
+        Args:
+            preset: string. A built in preset identifier, a Kaggle Models
+                handle, a Hugging Face handle, or a path to a local directory.
+            load_weights: bool. If `True`, the weights will be loaded into the
+                model architecture. If `False`, the weights will be randomly
+                initialized.
+
+        Examples:
+        ```python
+        # Load an audio converter from a preset.
+        converter = keras_nlp.layers.AudioConverter.from_preset(
+            "whisper_base_en"
+        )
+        # Convert some raw mono channel audio input.
+        converter(np.ones(2, 1_000))
+        ```
+        """
+        loader = get_preset_loader(preset)
+        backbone_cls = loader.check_backbone_class()
+        if cls.backbone_cls != backbone_cls:
+            cls = find_subclass(preset, cls, backbone_cls)
+        return loader.load_audio_converter(cls, **kwargs)
+
+    def save_to_preset(self, preset_dir):
+        """Save tokenizer to a preset directory.
+
+        Args:
+            preset_dir: The path to the local model preset directory.
+        """
+        save_serialized_object(
+            self,
+            preset_dir,
+            config_file=AUDIO_CONVERTER_CONFIG_FILE,
+        )
diff --git a/keras_nlp/src/layers/preprocessing/image_converter.py b/keras_nlp/src/layers/preprocessing/image_converter.py
new file mode 100644
index 0000000000..58ae3fb088
--- /dev/null
+++ b/keras_nlp/src/layers/preprocessing/image_converter.py
@@ -0,0 +1,150 @@
+# Copyright 2024 The KerasNLP Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import keras
+
+from keras_nlp.src.api_export import keras_nlp_export
+from keras_nlp.src.layers.preprocessing.preprocessing_layer import (
+    PreprocessingLayer,
+)
+from keras_nlp.src.utils.preset_utils import IMAGE_CONVERTER_CONFIG_FILE
+from keras_nlp.src.utils.preset_utils import find_subclass
+from keras_nlp.src.utils.preset_utils import get_preset_loader
+from keras_nlp.src.utils.preset_utils import save_serialized_object
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
+
+
+@keras_nlp_export("keras_nlp.layers.ImageConverter")
+class ImageConverter(PreprocessingLayer):
+    """Convert raw image for models that support image input.
+
+    This class converts from raw images of any size, to preprocessed
+    images for pretrained model inputs. It is meant to be a convenient way to
+    write custom preprocessing code that is not model specific. This layer
+    should be instantiated via the `from_preset()` constructor, which will
+    create the correct subclass of this layer for the model preset.
+
+    The layer will take a input a raw image tensor in the channels last or
+    channels first format, and output a preprocessed image input for modeling.
+    The exact structure of the output will vary per model, though in most cases
+    this layer will simply resize the image to the size needed by the model
+    input.
+
+    Examples:
+    ```python
+    # Resize images for `"pali_gemma_3b_224"`.
+    converter = keras_nlp.layers.ImageConverter.from_preset("pali_gemma_3b_224")
+    converter(np.ones(2, 512, 512, 3)) # Output shape: (2, 224, 224, 3)
+    # Resize images for `"pali_gemma_3b_224"`.
+    converter = keras_nlp.layers.ImageConverter.from_preset("pali_gemma_3b_448")
+    converter(np.ones(2, 512, 512, 3)) # Output shape: (2, 448, 448, 3)
+    ```
+    """
+
+    def __init__(
+        self,
+        height,
+        width,
+        crop_to_aspect_ratio=True,
+        pad_to_aspect_ratio=False,
+        interpolation="bilinear",
+        fill_mode="constant",
+        fill_value=0.0,
+        data_format=None,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        # By default, we just do a simple resize. Any model can subclass this
+        # layer for preprocessing of a raw image to a model image input.
+        self.resizing = keras.layers.Resizing(
+            height,
+            width,
+            crop_to_aspect_ratio=crop_to_aspect_ratio,
+            pad_to_aspect_ratio=pad_to_aspect_ratio,
+            interpolation=interpolation,
+            fill_mode=fill_mode,
+            fill_value=fill_value,
+            data_format=data_format,
+        )
+
+    @preprocessing_function
+    def call(self, inputs):
+        return self.resizing(inputs)
+
+    @classmethod
+    def from_preset(
+        cls,
+        preset,
+        **kwargs,
+    ):
+        """Instantiate a `keras_nlp.layers.ImageConverter` from a model preset.
+
+        A preset is a directory of configs, weights and other file assets used
+        to save and load a pre-trained model. The `preset` can be passed as a
+        one of:
+
+        1. a built in preset identifier like `'pali_gemma_3b_224'`
+        2. a Kaggle Models handle like
+           `'kaggle://user/paligemma/keras/pali_gemma_3b_224'`
+        3. a Hugging Face handle like `'hf://user/pali_gemma_3b_224'`
+        4. a path to a local preset directory like `'./pali_gemma_3b_224'`
+
+        You can run `cls.presets.keys()` to list all built-in presets available
+        on the class.
+
+        This constructor can be called in one of two ways. Either from the base
+        class like `keras_nlp.models.ImageConverter.from_preset()`, or from a
+        model class like
+        `keras_nlp.models.PaliGemmaImageConverter.from_preset()`. If calling
+        from the base class, the subclass of the returning object will be
+        inferred from the config in the preset directory.
+
+        Args:
+            preset: string. A built in preset identifier, a Kaggle Models
+                handle, a Hugging Face handle, or a path to a local directory.
+            load_weights: bool. If `True`, the weights will be loaded into the
+                model architecture. If `False`, the weights will be randomly
+                initialized.
+
+        Examples:
+        ```python
+        # Resize images for `"pali_gemma_3b_224"`.
+        converter = keras_nlp.layers.ImageConverter.from_preset(
+            "pali_gemma_3b_224"
+        )
+        converter(np.ones(2, 512, 512, 3)) # Output shape: (2, 224, 224, 3)
+        # Override arguments on the base class.
+        converter = keras_nlp.layers.ImageConverter.from_preset(
+            "pali_gemma_3b_448",
+            crop_to_aspect_ratio=False,
+        )
+        converter(np.ones(2, 512, 512, 3)) # (2, 448, 448, 3)
+        ```
+        """
+        loader = get_preset_loader(preset)
+        backbone_cls = loader.check_backbone_class()
+        if cls.backbone_cls != backbone_cls:
+            cls = find_subclass(preset, cls, backbone_cls)
+        return loader.load_image_converter(cls, **kwargs)
+
+    def save_to_preset(self, preset_dir):
+        """Save tokenizer to a preset directory.
+
+        Args:
+            preset_dir: The path to the local model preset directory.
+        """
+        save_serialized_object(
+            self,
+            preset_dir,
+            config_file=IMAGE_CONVERTER_CONFIG_FILE,
+        )
diff --git a/keras_nlp/src/layers/preprocessing/masked_lm_mask_generator.py b/keras_nlp/src/layers/preprocessing/masked_lm_mask_generator.py
index 2c14394437..fe78e0c172 100644
--- a/keras_nlp/src/layers/preprocessing/masked_lm_mask_generator.py
+++ b/keras_nlp/src/layers/preprocessing/masked_lm_mask_generator.py
@@ -18,7 +18,7 @@
     PreprocessingLayer,
 )
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow as tf
@@ -166,7 +166,7 @@ def __init__(
             random_token_rate=self.random_token_rate,
         )
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, inputs):
         inputs, unbatched, rectangular = convert_to_ragged_batch(inputs)
 
diff --git a/keras_nlp/src/layers/preprocessing/multi_segment_packer.py b/keras_nlp/src/layers/preprocessing/multi_segment_packer.py
index a972bf551c..53625783cc 100644
--- a/keras_nlp/src/layers/preprocessing/multi_segment_packer.py
+++ b/keras_nlp/src/layers/preprocessing/multi_segment_packer.py
@@ -17,7 +17,7 @@
     PreprocessingLayer,
 )
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow as tf
@@ -282,7 +282,7 @@ def _combine_inputs(
         segment_ids = tf.concat(segment_ids_to_combine, 1)
         return token_ids, segment_ids
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         inputs,
diff --git a/keras_nlp/src/layers/preprocessing/random_deletion.py b/keras_nlp/src/layers/preprocessing/random_deletion.py
index bd1b6764fa..6df8b6ef28 100644
--- a/keras_nlp/src/layers/preprocessing/random_deletion.py
+++ b/keras_nlp/src/layers/preprocessing/random_deletion.py
@@ -21,7 +21,7 @@
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
 from keras_nlp.src.utils.tensor_utils import is_int_dtype
 from keras_nlp.src.utils.tensor_utils import is_string_dtype
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow as tf
@@ -171,7 +171,7 @@ def __init__(
                 default_value=False,
             )
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, inputs):
         inputs, unbatched, rectangular = convert_to_ragged_batch(inputs)
 
diff --git a/keras_nlp/src/layers/preprocessing/random_swap.py b/keras_nlp/src/layers/preprocessing/random_swap.py
index 71b07193dc..32f7a08d26 100644
--- a/keras_nlp/src/layers/preprocessing/random_swap.py
+++ b/keras_nlp/src/layers/preprocessing/random_swap.py
@@ -21,7 +21,7 @@
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
 from keras_nlp.src.utils.tensor_utils import is_int_dtype
 from keras_nlp.src.utils.tensor_utils import is_string_dtype
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow as tf
@@ -167,7 +167,7 @@ def __init__(
                 default_value=False,
             )
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, inputs):
         inputs, unbatched, rectangular = convert_to_ragged_batch(inputs)
 
diff --git a/keras_nlp/src/layers/preprocessing/start_end_packer.py b/keras_nlp/src/layers/preprocessing/start_end_packer.py
index e1712eba91..a6e7b4d068 100644
--- a/keras_nlp/src/layers/preprocessing/start_end_packer.py
+++ b/keras_nlp/src/layers/preprocessing/start_end_packer.py
@@ -18,7 +18,7 @@
     PreprocessingLayer,
 )
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow as tf
@@ -155,7 +155,7 @@ def check_special_value_type(value, value_name):
         self.pad_value = pad_value
         self.return_padding_mask = return_padding_mask
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         inputs,
diff --git a/keras_nlp/src/models/bart/bart_preprocessor.py b/keras_nlp/src/models/bart/bart_preprocessor.py
index dc013779a3..0f9de65fab 100644
--- a/keras_nlp/src/models/bart/bart_preprocessor.py
+++ b/keras_nlp/src/models/bart/bart_preprocessor.py
@@ -20,7 +20,7 @@
 from keras_nlp.src.models.bart.bart_backbone import BartBackbone
 from keras_nlp.src.models.bart.bart_tokenizer import BartTokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.BartPreprocessor")
@@ -174,7 +174,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
diff --git a/keras_nlp/src/models/bloom/bloom_preprocessor.py b/keras_nlp/src/models/bloom/bloom_preprocessor.py
index 6c572591b7..8b2d7b2ba0 100644
--- a/keras_nlp/src/models/bloom/bloom_preprocessor.py
+++ b/keras_nlp/src/models/bloom/bloom_preprocessor.py
@@ -20,7 +20,7 @@
 from keras_nlp.src.models.bloom.bloom_backbone import BloomBackbone
 from keras_nlp.src.models.bloom.bloom_tokenizer import BloomTokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.BloomPreprocessor")
@@ -134,7 +134,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
diff --git a/keras_nlp/src/models/causal_lm_preprocessor.py b/keras_nlp/src/models/causal_lm_preprocessor.py
index 1713ce6566..817851493a 100644
--- a/keras_nlp/src/models/causal_lm_preprocessor.py
+++ b/keras_nlp/src/models/causal_lm_preprocessor.py
@@ -17,7 +17,7 @@
 from keras_nlp.src.layers.preprocessing.start_end_packer import StartEndPacker
 from keras_nlp.src.models.preprocessor import Preprocessor
 from keras_nlp.src.utils.tensor_utils import strip_to_ragged
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.CausalLMPreprocessor")
@@ -98,7 +98,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
@@ -124,7 +124,7 @@ def call(
         y, sample_weight = token_ids[..., 1:], padding_mask[..., 1:]
         return keras.utils.pack_x_y_sample_weight(x, y, sample_weight)
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def generate_preprocess(
         self,
         x,
@@ -153,7 +153,7 @@ def generate_preprocess(
             "padding_mask": padding_mask,
         }
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def generate_postprocess(
         self,
         x,
diff --git a/keras_nlp/src/models/deberta_v3/deberta_v3_masked_lm_preprocessor.py b/keras_nlp/src/models/deberta_v3/deberta_v3_masked_lm_preprocessor.py
index 8096e76e37..6823d1a8eb 100644
--- a/keras_nlp/src/models/deberta_v3/deberta_v3_masked_lm_preprocessor.py
+++ b/keras_nlp/src/models/deberta_v3/deberta_v3_masked_lm_preprocessor.py
@@ -22,7 +22,7 @@
     DebertaV3Tokenizer,
 )
 from keras_nlp.src.models.masked_lm_preprocessor import MaskedLMPreprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.DebertaV3MaskedLMPreprocessor")
@@ -119,7 +119,7 @@ class DebertaV3MaskedLMPreprocessor(MaskedLMPreprocessor):
     backbone_cls = DebertaV3Backbone
     tokenizer_cls = DebertaV3Tokenizer
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         output = super().call(x, y=y, sample_weight=sample_weight)
         x, y, sample_weight = keras.utils.unpack_x_y_sample_weight(output)
diff --git a/keras_nlp/src/models/deberta_v3/deberta_v3_text_classifier_preprocessor.py b/keras_nlp/src/models/deberta_v3/deberta_v3_text_classifier_preprocessor.py
index 6f1996d3b4..9634b77238 100644
--- a/keras_nlp/src/models/deberta_v3/deberta_v3_text_classifier_preprocessor.py
+++ b/keras_nlp/src/models/deberta_v3/deberta_v3_text_classifier_preprocessor.py
@@ -24,7 +24,7 @@
 from keras_nlp.src.models.text_classifier_preprocessor import (
     TextClassifierPreprocessor,
 )
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export(
@@ -156,7 +156,7 @@ class DebertaV3TextClassifierPreprocessor(TextClassifierPreprocessor):
     backbone_cls = DebertaV3Backbone
     tokenizer_cls = DebertaV3Tokenizer
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         output = super().call(x, y=y, sample_weight=sample_weight)
         x, y, sample_weight = keras.utils.unpack_x_y_sample_weight(output)
diff --git a/keras_nlp/src/models/distil_bert/distil_bert_masked_lm_preprocessor.py b/keras_nlp/src/models/distil_bert/distil_bert_masked_lm_preprocessor.py
index 9004fa06ba..1a69de4e50 100644
--- a/keras_nlp/src/models/distil_bert/distil_bert_masked_lm_preprocessor.py
+++ b/keras_nlp/src/models/distil_bert/distil_bert_masked_lm_preprocessor.py
@@ -22,7 +22,7 @@
     DistilBertTokenizer,
 )
 from keras_nlp.src.models.masked_lm_preprocessor import MaskedLMPreprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.DistilBertMaskedLMPreprocessor")
@@ -123,7 +123,7 @@ class DistilBertMaskedLMPreprocessor(MaskedLMPreprocessor):
     backbone_cls = DistilBertBackbone
     tokenizer_cls = DistilBertTokenizer
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         output = super().call(x, y=y, sample_weight=sample_weight)
         x, y, sample_weight = keras.utils.unpack_x_y_sample_weight(output)
diff --git a/keras_nlp/src/models/distil_bert/distil_bert_text_classifier_preprocessor.py b/keras_nlp/src/models/distil_bert/distil_bert_text_classifier_preprocessor.py
index 90ebe7ef3b..f4bda84a2c 100644
--- a/keras_nlp/src/models/distil_bert/distil_bert_text_classifier_preprocessor.py
+++ b/keras_nlp/src/models/distil_bert/distil_bert_text_classifier_preprocessor.py
@@ -25,7 +25,7 @@
 from keras_nlp.src.models.text_classifier_preprocessor import (
     TextClassifierPreprocessor,
 )
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export(
@@ -126,7 +126,7 @@ class DistilBertTextClassifierPreprocessor(TextClassifierPreprocessor):
     backbone_cls = DistilBertBackbone
     tokenizer_cls = DistilBertTokenizer
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         output = super().call(x, y=y, sample_weight=sample_weight)
         x, y, sample_weight = keras.utils.unpack_x_y_sample_weight(output)
diff --git a/keras_nlp/src/models/electra/electra_preprocessor.py b/keras_nlp/src/models/electra/electra_preprocessor.py
index 941e5d4d13..82dbb3310c 100644
--- a/keras_nlp/src/models/electra/electra_preprocessor.py
+++ b/keras_nlp/src/models/electra/electra_preprocessor.py
@@ -21,7 +21,7 @@
 from keras_nlp.src.models.electra.electra_backbone import ElectraBackbone
 from keras_nlp.src.models.electra.electra_tokenizer import ElectraTokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.ElectraPreprocessor")
@@ -142,7 +142,7 @@ def get_config(self):
         )
         return config
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         x = x if isinstance(x, tuple) else (x,)
         x = tuple(self.tokenizer(segment) for segment in x)
diff --git a/keras_nlp/src/models/f_net/f_net_masked_lm_preprocessor.py b/keras_nlp/src/models/f_net/f_net_masked_lm_preprocessor.py
index 7628f26ee4..3d0b625ef1 100644
--- a/keras_nlp/src/models/f_net/f_net_masked_lm_preprocessor.py
+++ b/keras_nlp/src/models/f_net/f_net_masked_lm_preprocessor.py
@@ -18,7 +18,7 @@
 from keras_nlp.src.models.f_net.f_net_backbone import FNetBackbone
 from keras_nlp.src.models.f_net.f_net_tokenizer import FNetTokenizer
 from keras_nlp.src.models.masked_lm_preprocessor import MaskedLMPreprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.FNetMaskedLMPreprocessor")
@@ -121,7 +121,7 @@ class FNetMaskedLMPreprocessor(MaskedLMPreprocessor):
     backbone_cls = FNetBackbone
     tokenizer_cls = FNetTokenizer
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         output = super().call(x, y=y, sample_weight=sample_weight)
         x, y, sample_weight = keras.utils.unpack_x_y_sample_weight(output)
diff --git a/keras_nlp/src/models/f_net/f_net_text_classifier_preprocessor.py b/keras_nlp/src/models/f_net/f_net_text_classifier_preprocessor.py
index c1f1f3f71e..124fc803e2 100644
--- a/keras_nlp/src/models/f_net/f_net_text_classifier_preprocessor.py
+++ b/keras_nlp/src/models/f_net/f_net_text_classifier_preprocessor.py
@@ -21,7 +21,7 @@
 from keras_nlp.src.models.text_classifier_preprocessor import (
     TextClassifierPreprocessor,
 )
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export(
@@ -124,7 +124,7 @@ class FNetTextClassifierPreprocessor(TextClassifierPreprocessor):
     backbone_cls = FNetBackbone
     tokenizer_cls = FNetTokenizer
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         # FNet has not padding mask.
         output = super().call(x, y=y, sample_weight=sample_weight)
diff --git a/keras_nlp/src/models/falcon/falcon_preprocessor.py b/keras_nlp/src/models/falcon/falcon_preprocessor.py
index 1c7fd3c138..491f6e5fe2 100644
--- a/keras_nlp/src/models/falcon/falcon_preprocessor.py
+++ b/keras_nlp/src/models/falcon/falcon_preprocessor.py
@@ -20,7 +20,7 @@
 from keras_nlp.src.models.falcon.falcon_backbone import FalconBackbone
 from keras_nlp.src.models.falcon.falcon_tokenizer import FalconTokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.FalconPreprocessor")
@@ -136,7 +136,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
diff --git a/keras_nlp/src/models/gemma/gemma_preprocessor.py b/keras_nlp/src/models/gemma/gemma_preprocessor.py
index 745f437ecd..dcbe531b56 100644
--- a/keras_nlp/src/models/gemma/gemma_preprocessor.py
+++ b/keras_nlp/src/models/gemma/gemma_preprocessor.py
@@ -20,7 +20,7 @@
 from keras_nlp.src.models.gemma.gemma_backbone import GemmaBackbone
 from keras_nlp.src.models.gemma.gemma_tokenizer import GemmaTokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.GemmaPreprocessor")
@@ -151,7 +151,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
diff --git a/keras_nlp/src/models/gpt2/gpt2_preprocessor.py b/keras_nlp/src/models/gpt2/gpt2_preprocessor.py
index b645ac7682..c9af92fcbf 100644
--- a/keras_nlp/src/models/gpt2/gpt2_preprocessor.py
+++ b/keras_nlp/src/models/gpt2/gpt2_preprocessor.py
@@ -20,7 +20,7 @@
 from keras_nlp.src.models.gpt2.gpt2_backbone import GPT2Backbone
 from keras_nlp.src.models.gpt2.gpt2_tokenizer import GPT2Tokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.GPT2Preprocessor")
@@ -136,7 +136,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
diff --git a/keras_nlp/src/models/gpt_neo_x/gpt_neo_x_preprocessor.py b/keras_nlp/src/models/gpt_neo_x/gpt_neo_x_preprocessor.py
index 25875d03da..06a46ce470 100644
--- a/keras_nlp/src/models/gpt_neo_x/gpt_neo_x_preprocessor.py
+++ b/keras_nlp/src/models/gpt_neo_x/gpt_neo_x_preprocessor.py
@@ -19,7 +19,7 @@
 from keras_nlp.src.models.gpt_neo_x.gpt_neo_x_backbone import GPTNeoXBackbone
 from keras_nlp.src.models.gpt_neo_x.gpt_neo_x_tokenizer import GPTNeoXTokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.GPTNeoXPreprocessor")
@@ -94,7 +94,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
diff --git a/keras_nlp/src/models/llama/llama_preprocessor.py b/keras_nlp/src/models/llama/llama_preprocessor.py
index 8b7c5772e6..75122856c6 100644
--- a/keras_nlp/src/models/llama/llama_preprocessor.py
+++ b/keras_nlp/src/models/llama/llama_preprocessor.py
@@ -18,7 +18,7 @@
 from keras_nlp.src.models.llama.llama_backbone import LlamaBackbone
 from keras_nlp.src.models.llama.llama_tokenizer import LlamaTokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.LlamaPreprocessor")
@@ -149,7 +149,7 @@ def get_config(self):
         )
         return config
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
diff --git a/keras_nlp/src/models/masked_lm_preprocessor.py b/keras_nlp/src/models/masked_lm_preprocessor.py
index 491b715280..c4dab122ab 100644
--- a/keras_nlp/src/models/masked_lm_preprocessor.py
+++ b/keras_nlp/src/models/masked_lm_preprocessor.py
@@ -21,7 +21,7 @@
     MultiSegmentPacker,
 )
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.MaskedLMPreprocessor")
@@ -113,7 +113,7 @@ def build(self, input_shape):
             unselectable_token_ids=self.tokenizer.special_token_ids,
         )
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         x = x if isinstance(x, tuple) else (x,)
         x = tuple(self.tokenizer(segment) for segment in x)
diff --git a/keras_nlp/src/models/mistral/mistral_preprocessor.py b/keras_nlp/src/models/mistral/mistral_preprocessor.py
index 0278103c54..c6d7731722 100644
--- a/keras_nlp/src/models/mistral/mistral_preprocessor.py
+++ b/keras_nlp/src/models/mistral/mistral_preprocessor.py
@@ -19,7 +19,7 @@
 from keras_nlp.src.models.mistral.mistral_backbone import MistralBackbone
 from keras_nlp.src.models.mistral.mistral_tokenizer import MistralTokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.MistralPreprocessor")
@@ -150,7 +150,7 @@ def get_config(self):
         )
         return config
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
diff --git a/keras_nlp/src/models/opt/opt_preprocessor.py b/keras_nlp/src/models/opt/opt_preprocessor.py
index 3b39c82cbc..0cafaaec5e 100644
--- a/keras_nlp/src/models/opt/opt_preprocessor.py
+++ b/keras_nlp/src/models/opt/opt_preprocessor.py
@@ -20,7 +20,7 @@
 from keras_nlp.src.models.opt.opt_backbone import OPTBackbone
 from keras_nlp.src.models.opt.opt_tokenizer import OPTTokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.OPTPreprocessor")
@@ -148,7 +148,7 @@ def get_config(self):
         )
         return config
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
diff --git a/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py b/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py
index be541597cd..4a899c6734 100644
--- a/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py
@@ -11,19 +11,13 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-import os
-
 import numpy as np
+import pytest
+from keras import ops
 
 from keras_nlp.src.models.pali_gemma.pali_gemma_backbone import (
     PaliGemmaBackbone,
 )
-from keras_nlp.src.models.pali_gemma.pali_gemma_causal_lm_preprocessor import (
-    PaliGemmaCausalLMPreprocessor,
-)
-from keras_nlp.src.models.pali_gemma.pali_gemma_tokenizer import (
-    PaliGemmaTokenizer,
-)
 from keras_nlp.src.tests.test_case import TestCase
 
 
@@ -34,15 +28,6 @@ def setUp(self):
         self.text_sequence_length = 64
         self.image_size = 16
         self.image_sequence_length = int((self.image_size / 4) ** 2)
-
-        proto = "gemma_test_vocab.spm"
-        tokenizer = PaliGemmaTokenizer(
-            os.path.join(self.get_test_data_dir(), proto)
-        )
-        self.preprocessor = PaliGemmaCausalLMPreprocessor(
-            tokenizer, self.text_sequence_length, False, False
-        )
-
         self.init_kwargs = {
             "vocabulary_size": self.vocabulary_size,
             "image_size": self.image_size,
@@ -65,7 +50,6 @@ def setUp(self):
         dummy_text_token_ids = np.random.rand(
             self.batch_size, self.text_sequence_length
         )
-        dummy_text = ["answer en the quick brown fox"] * self.batch_size
         self.input_data = {
             "token_ids": dummy_text_token_ids,
             "images": dummy_images,
@@ -78,11 +62,6 @@ def setUp(self):
                 dtype="int32",
             ),
         }
-        self.raw_input_data = {
-            "images": dummy_images,
-            "prompts": dummy_text,
-            "responses": dummy_text,
-        }
 
     def test_backbone_basics(self):
         self.run_backbone_test(
@@ -98,15 +77,37 @@ def test_backbone_basics(self):
             run_mixed_precision_check=False,  # TODO: Set to `True`
         )
 
-    def test_pali_gemma_backbone_with_preprocessing(self):
-        model = PaliGemmaBackbone(**self.init_kwargs)
-        x, _, _ = self.preprocessor(self.raw_input_data)
-        output = model(x)
-        self.assertEqual(
-            (
-                self.batch_size,
-                self.text_sequence_length + self.image_sequence_length,
-                8,
+    @pytest.mark.large
+    def test_saved_model(self):
+        self.run_model_saving_test(
+            cls=PaliGemmaBackbone,
+            init_kwargs=self.init_kwargs,
+            input_data=self.input_data,
+        )
+
+    @pytest.mark.large
+    def test_smallest_preset(self):
+        self.run_preset_test(
+            cls=PaliGemmaBackbone,
+            preset="pali_gemma_3b_mix_224",
+            input_data={
+                "token_ids": ops.array([[1169, 2068, 7586, 21831, 13]]),
+                "padding_mask": ops.ones((1, 5), dtype="int32"),
+                "response_mask": ops.zeros((1, 5), dtype="int32"),
+                "images": ops.zeros((1, 224, 224, 3), dtype="float32"),
+            },
+            expected_output_shape=(1, 261, 2048),
+            # The forward pass from a preset should be stable!
+            expected_partial_output=ops.array(
+                [-0.449851,  1.431027, -0.713446,  0.417485, -0.640859]
             ),
-            output.shape,
         )
+
+    @pytest.mark.extra_large
+    def test_all_presets(self):
+        for preset in PaliGemmaBackbone.presets:
+            self.run_preset_test(
+                cls=PaliGemmaBackbone,
+                preset=preset,
+                input_data=self.input_data,
+            )
diff --git a/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor.py b/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor.py
index 764b570fa2..3521f16073 100644
--- a/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor.py
@@ -17,22 +17,42 @@
 from keras_nlp.src.layers.preprocessing.multi_segment_packer import (
     MultiSegmentPacker,
 )
-from keras_nlp.src.models.gemma.gemma_causal_lm_preprocessor import (
-    GemmaCausalLMPreprocessor,
-)
+from keras_nlp.src.models.causal_lm_preprocessor import CausalLMPreprocessor
 from keras_nlp.src.models.pali_gemma.pali_gemma_backbone import (
     PaliGemmaBackbone,
 )
+from keras_nlp.src.models.pali_gemma.pali_gemma_image_converter import (
+    PaliGemmaImageConverter,
+)
 from keras_nlp.src.models.pali_gemma.pali_gemma_tokenizer import (
     PaliGemmaTokenizer,
 )
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.PaliGemmaCausalLMPreprocessor")
-class PaliGemmaCausalLMPreprocessor(GemmaCausalLMPreprocessor):
+class PaliGemmaCausalLMPreprocessor(CausalLMPreprocessor):
     backbone_cls = PaliGemmaBackbone
     tokenizer_cls = PaliGemmaTokenizer
+    image_converter_cls = PaliGemmaImageConverter
+
+    def __init__(
+        self,
+        tokenizer,
+        image_converter,
+        sequence_length=1024,
+        add_start_token=True,
+        add_end_token=True,
+        **kwargs,
+    ):
+        super().__init__(
+            tokenizer=tokenizer,
+            sequence_length=sequence_length,
+            add_start_token=add_start_token,
+            add_end_token=add_end_token,
+            **kwargs,
+        )
+        self.image_converter = image_converter
 
     def build(self, input_shape):
         # Defer packer creation to `build()` so that we can be sure tokenizer
@@ -46,7 +66,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
@@ -58,6 +78,8 @@ def call(
         images, prompts, responses = x["images"], x["prompts"], x["responses"]
         prompts = self.tokenizer(prompts)
         responses = self.tokenizer(responses)
+        if self.image_converter:
+            images = self.image_converter(images)
         # Pad with one extra token to account for the truncation below.
         token_ids, segment_ids = self.packer(
             (prompts, responses),
@@ -80,7 +102,7 @@ def call(
         sample_weight = response_mask[..., 1:]
         return keras.utils.pack_x_y_sample_weight(x, y, sample_weight)
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def generate_preprocess(
         self,
         x,
@@ -103,6 +125,8 @@ def generate_preprocess(
 
         images, prompts = x["images"], x["prompts"]
         prompts = self.tokenizer(prompts)
+        if self.image_converter:
+            images = self.image_converter(images)
         if "responses" in x:
             responses = self.tokenizer(x["responses"])
             segments = (prompts, responses)
diff --git a/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor_test.py b/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor_test.py
index cba98a9d97..5553bcd71a 100644
--- a/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor_test.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor_test.py
@@ -20,6 +20,9 @@
 from keras_nlp.src.models.pali_gemma.pali_gemma_causal_lm_preprocessor import (
     PaliGemmaCausalLMPreprocessor,
 )
+from keras_nlp.src.models.pali_gemma.pali_gemma_image_converter import (
+    PaliGemmaImageConverter,
+)
 from keras_nlp.src.models.pali_gemma.pali_gemma_tokenizer import (
     PaliGemmaTokenizer,
 )
@@ -34,14 +37,16 @@ def setUp(self):
                 self.get_test_data_dir(), "gemma_test_vocab.spm"
             ),
         )
+        self.image_converter = PaliGemmaImageConverter(width=224, height=224)
         self.init_kwargs = {
             "tokenizer": self.tokenizer,
+            "image_converter": self.image_converter,
             "sequence_length": 8,
         }
         self.input_data = {
             "prompts": ["the quick"],
             "responses": ["brown fox"],
-            "images": [np.zeros([1, 224, 224, 3])],
+            "images": [np.zeros([512, 512, 3])],
         }
 
     def test_preprocessor_basics(self):
@@ -54,7 +59,7 @@ def test_preprocessor_basics(self):
                     "token_ids": [[1, 4, 9, 5, 7, 2, 0, 0]],
                     "padding_mask": [[1, 1, 1, 1, 1, 1, 0, 0]],
                     "response_mask": [[0, 0, 0, 1, 1, 1, 0, 0]],
-                    "images": self.input_data["images"],
+                    "images": np.zeros([1, 224, 224, 3]),
                 },
                 [[4, 9, 5, 7, 2, 0, 0, 0]],  # Labels shifted.
                 [[0, 0, 1, 1, 1, 0, 0, 0]],  # Zero out unlabeled examples.
@@ -65,7 +70,7 @@ def test_no_start_end_token(self):
         input_data = {
             "prompts": ["the quick"] * 4,
             "responses": ["brown fox"] * 4,
-            "images": [np.zeros([1, 224, 224, 3])] * 4,
+            "images": [np.zeros([512, 512, 3])] * 4,
         }
         preprocessor = PaliGemmaCausalLMPreprocessor(
             **self.init_kwargs,
@@ -76,19 +81,21 @@ def test_no_start_end_token(self):
         self.assertAllEqual(x["token_ids"], [[4, 9, 5, 7, 0, 0, 0, 0]] * 4)
         self.assertAllEqual(x["padding_mask"], [[1, 1, 1, 1, 0, 0, 0, 0]] * 4)
         self.assertAllEqual(x["response_mask"], [[0, 0, 1, 1, 0, 0, 0, 0]] * 4)
+        self.assertAllEqual(x["images"], np.zeros([4, 224, 224, 3]))
         self.assertAllEqual(y, [[9, 5, 7, 0, 0, 0, 0, 0]] * 4)
         self.assertAllEqual(sw, [[0, 1, 1, 0, 0, 0, 0, 0]] * 4)
 
     def test_generate_preprocess(self):
         input_data = {
             "prompts": "the quick",
-            "images": np.zeros([1, 224, 224, 3]),
+            "images": np.zeros([1, 512, 512, 3]),
         }
         preprocessor = PaliGemmaCausalLMPreprocessor(**self.init_kwargs)
         x = preprocessor.generate_preprocess(input_data)
         self.assertAllEqual(x["token_ids"], [1, 4, 9, 0, 0, 0, 0, 0])
         self.assertAllEqual(x["padding_mask"], [1, 1, 1, 0, 0, 0, 0, 0])
         self.assertAllEqual(x["response_mask"], [0, 0, 0, 0, 0, 0, 0, 0])
+        self.assertAllEqual(x["images"], np.zeros([1, 224, 224, 3]))
 
     def test_generate_postprocess(self):
         input_data = {
diff --git a/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_test.py b/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_test.py
index 51afdc5fb7..04d0745a8f 100644
--- a/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_test.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_test.py
@@ -28,6 +28,9 @@
 from keras_nlp.src.models.pali_gemma.pali_gemma_tokenizer import (
     PaliGemmaTokenizer,
 )
+from keras_nlp.src.models.pali_gemma.pali_gemma_image_converter import (
+    PaliGemmaImageConverter,
+)
 from keras_nlp.src.tests.test_case import TestCase
 
 
@@ -39,17 +42,17 @@ def setUp(self):
         self.dummy_text = [
             "the quick brown fox" for _ in range(self.batch_size)
         ]
-        self.dummy_images = np.random.uniform(
-            size=(self.batch_size, self.image_size, self.image_size, 3)
-        )
+        self.dummy_images = np.random.uniform(size=(self.batch_size, 20, 20, 3))
 
         proto = "gemma_test_vocab.spm"
         tokenizer = PaliGemmaTokenizer(
             os.path.join(self.get_test_data_dir(), proto)
         )
+        image_converter = PaliGemmaImageConverter(16, 16)
         self.vocabulary_size = tokenizer.vocabulary_size()
         self.preprocessor = PaliGemmaCausalLMPreprocessor(
             tokenizer,
+            image_converter,
             self.text_sequence_length,
             add_start_token=False,
             add_end_token=False,
@@ -96,7 +99,9 @@ def test_saved_model(self):
             "token_ids": np.random.rand(
                 self.batch_size, self.text_sequence_length
             ),
-            "images": self.dummy_images,
+            "images": np.ones(
+                (self.batch_size, self.image_size, self.image_size, 3)
+            ),
             "padding_mask": np.ones(
                 (self.batch_size, self.text_sequence_length),
                 dtype="int32",
diff --git a/keras_nlp/api/utils/__init__.py b/keras_nlp/src/models/pali_gemma/pali_gemma_image_converter.py
similarity index 61%
rename from keras_nlp/api/utils/__init__.py
rename to keras_nlp/src/models/pali_gemma/pali_gemma_image_converter.py
index a904b48f59..21d95c9fa6 100644
--- a/keras_nlp/api/utils/__init__.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_image_converter.py
@@ -11,11 +11,13 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""DO NOT EDIT.
+from keras_nlp.src.api_export import keras_nlp_export
+from keras_nlp.src.layers.preprocessing.image_converter import ImageConverter
+from keras_nlp.src.models.pali_gemma.pali_gemma_backbone import (
+    PaliGemmaBackbone,
+)
 
-This file was autogenerated. Do not edit it by hand,
-since your modifications would be overwritten.
-"""
 
-from keras_nlp.src.utils.tensor_utils import convert_from_tf
-from keras_nlp.src.utils.tensor_utils import convert_to_tf
+@keras_nlp_export("keras_nlp.layers.PaliGemmaImageConverter")
+class PaliGemmaImageConverter(ImageConverter):
+    backbone_cls = PaliGemmaBackbone
diff --git a/keras_nlp/src/models/phi3/phi3_preprocessor.py b/keras_nlp/src/models/phi3/phi3_preprocessor.py
index caa5c9eab4..ce392b5088 100644
--- a/keras_nlp/src/models/phi3/phi3_preprocessor.py
+++ b/keras_nlp/src/models/phi3/phi3_preprocessor.py
@@ -18,7 +18,7 @@
 from keras_nlp.src.models.phi3.phi3_backbone import Phi3Backbone
 from keras_nlp.src.models.phi3.phi3_tokenizer import Phi3Tokenizer
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.Phi3Preprocessor")
@@ -150,7 +150,7 @@ def get_config(self):
         )
         return config
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
diff --git a/keras_nlp/src/models/preprocessor.py b/keras_nlp/src/models/preprocessor.py
index e498746685..ef1ce6d503 100644
--- a/keras_nlp/src/models/preprocessor.py
+++ b/keras_nlp/src/models/preprocessor.py
@@ -48,6 +48,8 @@ class Preprocessor(PreprocessingLayer):
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
         self._tokenizer = None
+        self._image_converter = None
+        self._audio_converter = None
 
     def __setattr__(self, name, value):
         # Work around torch setattr for properties.
@@ -64,15 +66,54 @@ def tokenizer(self):
     def tokenizer(self, value):
         self._tokenizer = value
 
+    @property
+    def audio_converter(self):
+        """The audio converter used to preprocess audio data."""
+        return self._audio_converter
+
+    @audio_converter.setter
+    def audio_converter(self, value):
+        self._audio_converter = value
+
+    @property
+    def image_converter(self):
+        """The image converter used to preprocess image data."""
+        return self._image_converter
+
+    @image_converter.setter
+    def image_converter(self, value):
+        self._image_converter = value
+
     def get_config(self):
         config = super().get_config()
-        config["tokenizer"] = keras.layers.serialize(self.tokenizer)
+        if self.tokenizer:
+            config["tokenizer"] = keras.layers.serialize(self.tokenizer)
+        if self.audio_converter:
+            config["audio_converter"] = keras.layers.serialize(
+                self.audio_converter
+            )
+        if self.image_converter:
+            config["image_converter"] = keras.layers.serialize(
+                self.image_converter
+            )
         return config
 
     @classmethod
     def from_config(cls, config):
         if "tokenizer" in config and isinstance(config["tokenizer"], dict):
             config["tokenizer"] = keras.layers.deserialize(config["tokenizer"])
+        if "audio_converter" in config and isinstance(
+            config["audio_converter"], dict
+        ):
+            config["audio_converter"] = keras.layers.deserialize(
+                config["audio_converter"]
+            )
+        if "image_converter" in config and isinstance(
+            config["image_converter"], dict
+        ):
+            config["image_converter"] = keras.layers.deserialize(
+                config["image_converter"]
+            )
         return cls(**config)
 
     @classproperty
@@ -155,4 +196,9 @@ def save_to_preset(self, preset_dir):
             preset_dir,
             config_file=PREPROCESSOR_CONFIG_FILE,
         )
-        self.tokenizer.save_to_preset(preset_dir)
+        if self.tokenizer:
+            self.tokenizer.save_to_preset(preset_dir)
+        if self.audio_converter:
+            self.audio_converter.save_to_preset(preset_dir)
+        if self.image_converter:
+            self.image_converter.save_to_preset(preset_dir)
diff --git a/keras_nlp/src/models/roberta/roberta_masked_lm_preprocessor.py b/keras_nlp/src/models/roberta/roberta_masked_lm_preprocessor.py
index 4b3bf02b6c..369a34ce61 100644
--- a/keras_nlp/src/models/roberta/roberta_masked_lm_preprocessor.py
+++ b/keras_nlp/src/models/roberta/roberta_masked_lm_preprocessor.py
@@ -21,7 +21,7 @@
 from keras_nlp.src.models.masked_lm_preprocessor import MaskedLMPreprocessor
 from keras_nlp.src.models.roberta.roberta_backbone import RobertaBackbone
 from keras_nlp.src.models.roberta.roberta_tokenizer import RobertaTokenizer
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.RobertaMaskedLMPreprocessor")
@@ -137,7 +137,7 @@ def build(self, input_shape):
             sequence_length=self.sequence_length,
         )
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         output = super().call(x, y=y, sample_weight=sample_weight)
         x, y, sample_weight = keras.utils.unpack_x_y_sample_weight(output)
diff --git a/keras_nlp/src/models/roberta/roberta_text_classifier_preprocessor.py b/keras_nlp/src/models/roberta/roberta_text_classifier_preprocessor.py
index 377cc2cb75..a905156615 100644
--- a/keras_nlp/src/models/roberta/roberta_text_classifier_preprocessor.py
+++ b/keras_nlp/src/models/roberta/roberta_text_classifier_preprocessor.py
@@ -23,7 +23,7 @@
 from keras_nlp.src.models.text_classifier_preprocessor import (
     TextClassifierPreprocessor,
 )
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export(
@@ -151,7 +151,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         output = super().call(x, y=y, sample_weight=sample_weight)
         x, y, sample_weight = keras.utils.unpack_x_y_sample_weight(output)
diff --git a/keras_nlp/src/models/seq_2_seq_lm_preprocessor.py b/keras_nlp/src/models/seq_2_seq_lm_preprocessor.py
index 8da697dc99..84f49f3b90 100644
--- a/keras_nlp/src/models/seq_2_seq_lm_preprocessor.py
+++ b/keras_nlp/src/models/seq_2_seq_lm_preprocessor.py
@@ -17,7 +17,7 @@
 from keras_nlp.src.layers.preprocessing.start_end_packer import StartEndPacker
 from keras_nlp.src.models.preprocessor import Preprocessor
 from keras_nlp.src.utils.tensor_utils import strip_to_ragged
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow as tf
@@ -114,7 +114,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(
         self,
         x,
@@ -154,7 +154,7 @@ def call(
         sample_weight = decoder_padding_mask[..., 1:]
         return keras.utils.pack_x_y_sample_weight(x, y, sample_weight)
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def generate_preprocess(
         self,
         x,
@@ -215,7 +215,7 @@ def generate_preprocess(
             "decoder_padding_mask": decoder_padding_mask,
         }
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def generate_postprocess(
         self,
         x,
diff --git a/keras_nlp/src/models/text_classifier_preprocessor.py b/keras_nlp/src/models/text_classifier_preprocessor.py
index 6b8639d045..9b34353938 100644
--- a/keras_nlp/src/models/text_classifier_preprocessor.py
+++ b/keras_nlp/src/models/text_classifier_preprocessor.py
@@ -18,7 +18,7 @@
     MultiSegmentPacker,
 )
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.TextClassifierPreprocessor")
@@ -104,7 +104,7 @@ def build(self, input_shape):
             sequence_length=self.sequence_length,
         )
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         x = x if isinstance(x, tuple) else (x,)
         x = tuple(self.tokenizer(segment) for segment in x)
diff --git a/keras_nlp/src/models/whisper/whisper_audio_feature_extractor.py b/keras_nlp/src/models/whisper/whisper_audio_converter.py
similarity index 92%
rename from keras_nlp/src/models/whisper/whisper_audio_feature_extractor.py
rename to keras_nlp/src/models/whisper/whisper_audio_converter.py
index 0fdc4374e0..8adae7a30a 100644
--- a/keras_nlp/src/models/whisper/whisper_audio_feature_extractor.py
+++ b/keras_nlp/src/models/whisper/whisper_audio_converter.py
@@ -15,24 +15,18 @@
 
 import numpy as np
 
+from keras_nlp.src.api_export import keras_nlp_export
+from keras_nlp.src.layers.preprocessing.audio_converter import AudioConverter
+
 try:
     import tensorflow as tf
 except ImportError:
-    raise ImportError(
-        "To use `keras_nlp`, please install Tensorflow: `pip install tensorflow`. "
-        "The TensorFlow package is required for data preprocessing with any backend."
-    )
-
-from keras_nlp.src.api_export import keras_nlp_export
-from keras_nlp.src.layers.preprocessing.preprocessing_layer import (
-    PreprocessingLayer,
-)
+    tf = None
 
 
-@keras_nlp_export("keras_nlp.models.WhisperAudioFeatureExtractor")
-class WhisperAudioFeatureExtractor(PreprocessingLayer):
-    """
-    Whisper audio feature extractor layer.
+@keras_nlp_export("keras_nlp.models.WhisperAudioConverter")
+class WhisperAudioConverter(AudioConverter):
+    """Whisper audio converter layer.
 
     This layer takes in a batch of audio tensors, and computes the log-mel
     spectrogram features for each audio tensor.
@@ -55,19 +49,20 @@ class WhisperAudioFeatureExtractor(PreprocessingLayer):
             `max_audio_length * sampling_rate`. Defaults to `30`.
 
     Examples:
-
     ```python
     audio_tensor = tf.ones((8000,), dtype="float32")
 
     # Compute the log-mel spectrogram.
-    whisper_audio_feature_extractor = keras_nlp.models.WhisperAudioFeatureExtractor()
-    whisper_audio_feature_extractor(audio_tensor)
+    audio_converter = keras_nlp.models.WhisperAudioConverter.from_preset(
+        "whisper_base_en",
+    )
+    audio_converter(audio_tensor)
 
     # Compute the log-mel spectrogram for a batch of audio tensors.
     audio_tensor_1 = tf.ones((8000,), dtype="float32")
-    audio_tensor_2 = tf.ones((10000,), dtype="float32"
+    audio_tensor_2 = tf.ones((10000,), dtype="float32")
     audio_tensor = tf.ragged.stack([audio_tensor_1, audio_tensor_2], axis=0)
-    whisper_audio_feature_extractor(audio_tensor)
+    audio_converter(audio_tensor)
     ```
     """
 
diff --git a/keras_nlp/src/models/whisper/whisper_audio_feature_extractor_test.py b/keras_nlp/src/models/whisper/whisper_audio_converter_test.py
similarity index 84%
rename from keras_nlp/src/models/whisper/whisper_audio_feature_extractor_test.py
rename to keras_nlp/src/models/whisper/whisper_audio_converter_test.py
index 823acbf3d1..16923787f4 100644
--- a/keras_nlp/src/models/whisper/whisper_audio_feature_extractor_test.py
+++ b/keras_nlp/src/models/whisper/whisper_audio_converter_test.py
@@ -14,13 +14,13 @@
 
 import tensorflow as tf
 
-from keras_nlp.src.models.whisper.whisper_audio_feature_extractor import (
-    WhisperAudioFeatureExtractor,
+from keras_nlp.src.models.whisper.whisper_audio_converter import (
+    WhisperAudioConverter,
 )
 from keras_nlp.src.tests.test_case import TestCase
 
 
-class WhisperAudioFeatureExtractorTest(TestCase):
+class WhisperAudioConverterTest(TestCase):
     def setUp(self):
         self.init_kwargs = {
             "num_mels": 80,
@@ -38,14 +38,14 @@ def setUp(self):
 
     def test_feature_extractor_basics(self):
         self.run_preprocessing_layer_test(
-            cls=WhisperAudioFeatureExtractor,
+            cls=WhisperAudioConverter,
             init_kwargs=self.init_kwargs,
             input_data=self.input_data,
         )
 
     def test_correctness(self):
         audio_tensor = tf.ones((2,), dtype="float32")
-        outputs = WhisperAudioFeatureExtractor(**self.init_kwargs)(audio_tensor)
+        outputs = WhisperAudioConverter(**self.init_kwargs)(audio_tensor)
 
         # Verify shape.
         self.assertEqual(outputs.shape, (5, 80))
diff --git a/keras_nlp/src/models/whisper/whisper_preprocessor.py b/keras_nlp/src/models/whisper/whisper_preprocessor.py
deleted file mode 100644
index 8a65e6d004..0000000000
--- a/keras_nlp/src/models/whisper/whisper_preprocessor.py
+++ /dev/null
@@ -1,315 +0,0 @@
-# Copyright 2024 The KerasNLP Authors
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     https://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-
-import keras
-from absl import logging
-
-from keras_nlp.src.api_export import keras_nlp_export
-from keras_nlp.src.layers.preprocessing.start_end_packer import StartEndPacker
-from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.models.whisper.whisper_audio_feature_extractor import (
-    WhisperAudioFeatureExtractor,
-)
-from keras_nlp.src.models.whisper.whisper_backbone import WhisperBackbone
-from keras_nlp.src.models.whisper.whisper_tokenizer import WhisperTokenizer
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
-
-
-@keras_nlp_export("keras_nlp.models.WhisperPreprocessor")
-class WhisperPreprocessor(Preprocessor):
-    """A Whisper preprocessing layer which handles audio and text input.
-
-    This preprocessing layer will do three things:
-
-     1. Compute the log-mel spectrogram of the audio tensor inputs using
-        `audio_feature_extractor`.
-     2. Tokenize decoder inputs using the `tokenizer`.
-     2. Add the appropriate special tokens - `"<|startoftranscript|>", task
-        token, language token, `"<|endoftext|>"`, etc.
-     3. Construct a dictionary with keys `"encoder_features"`,
-        `"decoder_token_ids"`, `"decoder_padding_mask"` that can be passed
-        directly to a Whisper model.
-
-    Args:
-        tokenizer: A `keras_nlp.models.WhisperTokenizer` instance.
-        audio_feature_extractor: A
-            `keras_nlp.models.WhisperAudioFeatureExtractor` instance or `None`.
-            If `None` a feature extractor with default parameters will be
-            created.
-        decoder_sequence_length: The length of the packed decoder inputs.
-        language: string, language token. Should only be passed if your
-            tokenizer is multilingual.
-        task: string, task name. One of `"transcribe"`, `"translate"`. Should
-            only be passed if your tokenizer is multilingual.
-        no_timestamps: bool. If True, `"<|no_timestamps|>"` will be added as a
-            special token to your input.
-
-    Call arguments:
-        x: A dictionary with `"encoder_audio"` and `"decoder_text"` as its keys.
-            `"encoder_audio"` should correspond to the input audio tensor.
-            `"decoder_text"` should be a tensor of single string sequences.
-            Inputs may be batched or unbatched. Raw python inputs will be
-            converted to tensors.
-        y: Any label data. Will be passed through unaltered.
-        sample_weight: Any label weight data. Will be passed through unaltered.
-
-    Examples:
-
-    Directly calling the layer on data.
-    ```python
-    preprocessor = keras_nlp.models.WhisperPreprocessor.from_preset(
-        "whisper_tiny_en",
-    )
-
-    # Preprocess unbatched inputs.
-    input_data = {
-        "encoder_audio": tf.ones((200,)),
-        "decoder_text": "The quick brown fox jumped.",
-    }
-    preprocessor(input_data)
-
-    # Preprocess batched inputs.
-    input_data = {
-        "encoder_audio": tf.ones((2, 200)),
-        "decoder_text": ["The quick brown fox jumped.", "Call me Ishmael."],
-    }
-    preprocessor(input_data)
-
-    # Custom audio feature extractor and vocabulary.
-    audio_feature_extractor = keras_nlp.models.WhisperAudioFeatureExtractor(
-        num_mels=80,
-        num_fft_bins=400,
-        stride=100,
-        sampling_rate=100,
-        max_audio_length=5,
-    )
-
-    features = ["a quick fox.", "a fox quick."]
-    vocab = {"<|endoftext|>": 0, "a": 4, "Ġquick": 5, "Ġfox": 6}
-    merges = ["Ġ q", "u i", "c k", "ui ck", "Ġq uick"]
-    merges += ["Ġ f", "o x", "Ġf ox"]
-    special_tokens = {
-        "<|startoftranscript|>": 9,
-        "<|endoftext|>": 10,
-        "<|notimestamps|>": 11,
-        "<|transcribe|>": 12,
-        "<|translate|>": 13,
-    }
-
-    tokenizer = keras_nlp.models.WhisperTokenizer(
-        vocabulary=vocab,
-        merges=merges,
-        special_tokens=special_tokens,
-    )
-    preprocessor = keras_nlp.models.WhisperPreprocessor(
-        audio_feature_extractor=audio_feature_extractor,
-        tokenizer=tokenizer,
-    )
-
-    input_data = {
-        "encoder_audio": tf.ones((200,)),
-        "decoder_text": "The quick brown fox jumped.",
-    }
-    preprocessor(input_data)
-    ```
-
-    Mapping with `tf.data.Dataset`.
-    ```python
-    preprocessor = keras_nlp.models.WhisperPreprocessor.from_preset(
-        "whisper_tiny_en")
-
-    # Map labeled single sentences.
-    features = {
-        "encoder_audio": tf.ones((2, 200)),
-        "decoder_text": ["The quick brown fox jumped.", "Call me Ishmael."],
-    }
-    labels = tf.constant(["True", "False"])
-    ds = tf.data.Dataset.from_tensor_slices((features, labels))
-    ds = ds.map(preprocessor, num_parallel_calls=tf.data.AUTOTUNE)
-
-    # Map unlabeled single sentences.
-    features = {
-        "encoder_audio": tf.ones((2, 200)),
-        "decoder_text": ["The quick brown fox jumped.", "Call me Ishmael."],
-    }
-    ds = tf.data.Dataset.from_tensor_slices(features)
-    ds = ds.map(preprocessor, num_parallel_calls=tf.data.AUTOTUNE)
-    ```
-    """
-
-    backbone_cls = WhisperBackbone
-    tokenizer_cls = WhisperTokenizer
-
-    def __init__(
-        self,
-        tokenizer,
-        audio_feature_extractor=None,
-        decoder_sequence_length=448,
-        language=None,
-        task=None,
-        no_timestamps=True,
-        **kwargs,
-    ):
-        super().__init__(**kwargs)
-        if audio_feature_extractor is None:
-            audio_feature_extractor = WhisperAudioFeatureExtractor()
-        self.audio_feature_extractor = audio_feature_extractor
-        self.tokenizer = tokenizer
-        self.decoder_packer = None
-        self.decoder_sequence_length = decoder_sequence_length
-        self.language = language
-        self.task = task
-        self.no_timestamps = no_timestamps
-
-    def build(self, input_shape):
-        # Defer packer creation to `build()` so that we can be sure tokenizer
-        # assets have loaded when restoring a saved model.
-
-        # Create list of tokens to be prepended to decoder inputs.
-        bos_tokens = [self.tokenizer.bos_token_id]
-        if self.tokenizer.language_tokens is not None:
-            if (
-                self.language is None
-                or self.language not in self.tokenizer.language_tokens
-            ):
-                raise ValueError(
-                    "You must pass a non-None value for `language` when using "
-                    "a multilingual tokenizer. The value must be one of "
-                    f'{",".join(self.tokenizer.language_tokens.keys())}. '
-                    f"Received: language={self.language}."
-                )
-            if self.task is None or self.task not in [
-                "transcribe",
-                "translate",
-            ]:
-                raise ValueError(
-                    "You must pass a non-None value for `task` when using "
-                    "a multilingual tokenizer. The value must be one of "
-                    '`"transcribe"`, `"translate"`. '
-                    f"Received: task={self.task}."
-                )
-
-            bos_tokens += [self.tokenizer.language_tokens[self.language]]
-
-            special_token_dict = self.tokenizer._special_token_dict
-            if self.task == "transcribe":
-                bos_tokens += [special_token_dict["<|transcribe|>"]]
-            elif self.task == "translate":
-                bos_tokens += [special_token_dict["<|translate|>"]]
-        else:
-            if self.language is not None:
-                logging.info(
-                    "`tokenizer` is monolingual, and `language` has a "
-                    "non-`None` value. Setting `language` to `None`."
-                )
-                self.language = None
-            if self.task is not None:
-                logging.info(
-                    "`tokenizer` is monolingual, and `task` has a "
-                    "non-`None` value. Setting `task` to `None`."
-                )
-                self.task = None
-
-        if self.no_timestamps:
-            bos_tokens += [self.tokenizer.no_timestamps_token_id]
-
-        # TODO: Use `MultiSegmentPacker` instead of `StartEndPacker` once we
-        # want to move to multi-segment packing and have improved
-        # `MultiSegmentPacker`'s performance.
-        self.decoder_packer = StartEndPacker(
-            start_value=bos_tokens,
-            end_value=self.tokenizer.eos_token_id,
-            pad_value=self.tokenizer.pad_token_id,
-            sequence_length=self.decoder_sequence_length,
-            return_padding_mask=True,
-        )
-
-    @tf_preprocessing_function
-    def call(self, x, y=None, sample_weight=None, decoder_sequence_length=None):
-        if not (
-            isinstance(x, dict)
-            and ["encoder_audio", "decoder_text"] == list(x.keys())
-        ):
-            raise ValueError(
-                '`x` must be a dictionary, containing the keys `"encoder_audio"`'
-                f' and `"decoder_text"`. Received x={x}.'
-            )
-
-        encoder_features = self.audio_feature_extractor(x["encoder_audio"])
-        decoder_sequence_length = (
-            decoder_sequence_length or self.decoder_sequence_length
-        )
-        decoder_inputs = self.tokenizer(x["decoder_text"])
-        decoder_token_ids, decoder_padding_mask = self.decoder_packer(
-            decoder_inputs,
-            sequence_length=decoder_sequence_length,
-        )
-
-        x = {
-            "encoder_features": encoder_features,
-            "decoder_token_ids": decoder_token_ids,
-            "decoder_padding_mask": decoder_padding_mask,
-        }
-
-        return keras.utils.pack_x_y_sample_weight(x, y, sample_weight)
-
-    def get_config(self):
-        config = super().get_config()
-        config.update(
-            {
-                "audio_feature_extractor": keras.layers.serialize(
-                    self.audio_feature_extractor
-                ),
-                "decoder_sequence_length": self.decoder_sequence_length,
-                "language": self.language,
-                "task": self.task,
-                "no_timestamps": self.no_timestamps,
-            }
-        )
-        return config
-
-    @classmethod
-    def from_config(cls, config):
-        if "tokenizer" in config and isinstance(config["tokenizer"], dict):
-            config["tokenizer"] = keras.layers.deserialize(config["tokenizer"])
-
-        if "audio_feature_extractor" in config and isinstance(
-            config["audio_feature_extractor"], dict
-        ):
-            config["audio_feature_extractor"] = keras.layers.deserialize(
-                config["audio_feature_extractor"]
-            )
-
-        return cls(**config)
-
-    @property
-    def decoder_sequence_length(self):
-        """The padded length of decoder input sequences."""
-        return self._decoder_sequence_length
-
-    @decoder_sequence_length.setter
-    def decoder_sequence_length(self, value):
-        self._decoder_sequence_length = value
-        if self.decoder_packer is not None:
-            self.decoder_packer.sequence_length = value
-
-    @property
-    def sequence_length(self):
-        """Alias for `decoder_sequence_length`."""
-        return self.decoder_sequence_length
-
-    @sequence_length.setter
-    def sequence_length(self, value):
-        self.decoder_sequence_length = value
diff --git a/keras_nlp/src/models/whisper/whisper_preprocessor_test.py b/keras_nlp/src/models/whisper/whisper_preprocessor_test.py
deleted file mode 100644
index 9721ddcad6..0000000000
--- a/keras_nlp/src/models/whisper/whisper_preprocessor_test.py
+++ /dev/null
@@ -1,85 +0,0 @@
-# Copyright 2024 The KerasNLP Authors
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     https://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import numpy as np
-
-from keras_nlp.src.models.whisper.whisper_audio_feature_extractor import (
-    WhisperAudioFeatureExtractor,
-)
-from keras_nlp.src.models.whisper.whisper_preprocessor import (
-    WhisperPreprocessor,
-)
-from keras_nlp.src.models.whisper.whisper_tokenizer import WhisperTokenizer
-from keras_nlp.src.tests.test_case import TestCase
-
-
-class WhisperPreprocessorTest(TestCase):
-    def setUp(self):
-        self.audio_feature_extractor = WhisperAudioFeatureExtractor(
-            num_mels=80,
-            num_fft_bins=400,
-            stride=100,
-            sampling_rate=100,
-            max_audio_length=5,
-        )
-        self.vocab = ["air", "Ġair", "plane", "Ġat", "port"]
-        self.vocab = dict([(token, i) for i, token in enumerate(self.vocab)])
-        self.merges = ["Ġ a", "Ġ t", "Ġ i", "Ġ b", "a i", "p l", "n e"]
-        self.merges += ["Ġa t", "p o", "r t", "Ġt h", "ai r", "pl a", "po rt"]
-        self.merges += ["Ġai r", "Ġa i", "pla ne"]
-        self.special_tokens = {
-            "<|startoftranscript|>": 9,
-            "<|endoftext|>": 10,
-            "<|notimestamps|>": 11,
-            "<|transcribe|>": 12,
-            "<|translate|>": 13,
-        }
-        self.language_tokens = {
-            "<|en|>": 14,
-            "<|fr|>": 15,
-        }
-        self.tokenizer = WhisperTokenizer(
-            vocabulary=self.vocab,
-            merges=self.merges,
-            special_tokens=self.special_tokens,
-            language_tokens=self.language_tokens,
-        )
-        self.init_kwargs = {
-            "audio_feature_extractor": self.audio_feature_extractor,
-            "tokenizer": self.tokenizer,
-            "decoder_sequence_length": 12,
-            "language": "<|en|>",
-            "task": "translate",
-        }
-        self.input_data = {
-            "encoder_audio": np.ones((2, 200)),
-            "decoder_text": [" airplane at airport", " airplane at"],
-        }
-
-    def test_feature_extractor_basics(self):
-        self.run_preprocessor_test(
-            cls=WhisperPreprocessor,
-            init_kwargs=self.init_kwargs,
-            input_data=self.input_data,
-            token_id_key="decoder_token_ids",
-        )
-
-    def test_sequence_length_override(self):
-        input_data = {
-            "encoder_audio": np.ones((200,)),
-            "decoder_text": " airplane at airport",
-        }
-        preprocessor = WhisperPreprocessor(**self.init_kwargs)
-        x = preprocessor(input_data, decoder_sequence_length=6)
-        self.assertAllEqual(x["decoder_token_ids"], [9, 14, 13, 11, 1, 10])
diff --git a/keras_nlp/src/models/xlm_roberta/xlm_roberta_masked_lm_preprocessor.py b/keras_nlp/src/models/xlm_roberta/xlm_roberta_masked_lm_preprocessor.py
index 8feb7d674f..d1f028aac4 100644
--- a/keras_nlp/src/models/xlm_roberta/xlm_roberta_masked_lm_preprocessor.py
+++ b/keras_nlp/src/models/xlm_roberta/xlm_roberta_masked_lm_preprocessor.py
@@ -25,7 +25,7 @@
 from keras_nlp.src.models.xlm_roberta.xlm_roberta_tokenizer import (
     XLMRobertaTokenizer,
 )
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export("keras_nlp.models.XLMRobertaMaskedLMPreprocessor")
@@ -140,7 +140,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         output = super().call(x, y=y, sample_weight=sample_weight)
         x, y, sample_weight = keras.utils.unpack_x_y_sample_weight(output)
diff --git a/keras_nlp/src/models/xlm_roberta/xlm_roberta_text_classifier_preprocessor.py b/keras_nlp/src/models/xlm_roberta/xlm_roberta_text_classifier_preprocessor.py
index 756b935dd0..52984e9ad7 100644
--- a/keras_nlp/src/models/xlm_roberta/xlm_roberta_text_classifier_preprocessor.py
+++ b/keras_nlp/src/models/xlm_roberta/xlm_roberta_text_classifier_preprocessor.py
@@ -28,7 +28,7 @@
 from keras_nlp.src.models.xlm_roberta.xlm_roberta_tokenizer import (
     XLMRobertaTokenizer,
 )
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export(
@@ -169,7 +169,7 @@ def build(self, input_shape):
         )
         self.built = True
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, x, y=None, sample_weight=None):
         output = super().call(x, y=y, sample_weight=sample_weight)
         x, y, sample_weight = keras.utils.unpack_x_y_sample_weight(output)
diff --git a/keras_nlp/src/tokenizers/byte_pair_tokenizer.py b/keras_nlp/src/tokenizers/byte_pair_tokenizer.py
index 26096f5ec4..5eecf4cbfc 100644
--- a/keras_nlp/src/tokenizers/byte_pair_tokenizer.py
+++ b/keras_nlp/src/tokenizers/byte_pair_tokenizer.py
@@ -31,7 +31,7 @@
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
 from keras_nlp.src.utils.tensor_utils import is_int_dtype
 from keras_nlp.src.utils.tensor_utils import is_string_dtype
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow as tf
@@ -534,7 +534,7 @@ def _check_vocabulary(self):
                 "layer."
             )
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def tokenize(self, inputs):
         self._check_vocabulary()
         if self.add_prefix_space:
@@ -598,7 +598,7 @@ def process_unseen_tokens():
 
         return tokens
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def detokenize(self, inputs):
         self._check_vocabulary()
         inputs, unbatched, rectangular = convert_to_ragged_batch(inputs)
diff --git a/keras_nlp/src/tokenizers/byte_tokenizer.py b/keras_nlp/src/tokenizers/byte_tokenizer.py
index 594b2c2ffc..6b70df35ce 100644
--- a/keras_nlp/src/tokenizers/byte_tokenizer.py
+++ b/keras_nlp/src/tokenizers/byte_tokenizer.py
@@ -26,7 +26,7 @@
 from keras_nlp.src.tokenizers import tokenizer
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
 from keras_nlp.src.utils.tensor_utils import is_int_dtype
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow_text as tf_text
@@ -212,7 +212,7 @@ def get_vocabulary(self):
             vocab[chr(i)] = i
         return vocab
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def tokenize(self, inputs):
         unbatched = inputs.shape.rank == 0
         if unbatched:
@@ -243,7 +243,7 @@ def tokenize(self, inputs):
             tokens = tf.squeeze(tokens, 0)
         return tokens
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def detokenize(self, inputs):
         inputs, unbatched, rectangular = convert_to_ragged_batch(inputs)
         # Remove trailing padding tokens, so that trailing "\x00" bytes don't
diff --git a/keras_nlp/src/tokenizers/sentence_piece_tokenizer.py b/keras_nlp/src/tokenizers/sentence_piece_tokenizer.py
index 37d609128b..31ed3401ca 100644
--- a/keras_nlp/src/tokenizers/sentence_piece_tokenizer.py
+++ b/keras_nlp/src/tokenizers/sentence_piece_tokenizer.py
@@ -32,7 +32,7 @@
 from keras_nlp.src.utils.tensor_utils import is_int_dtype
 from keras_nlp.src.utils.tensor_utils import is_string_dtype
 from keras_nlp.src.utils.tensor_utils import tensor_to_list
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow_text as tf_text
@@ -235,7 +235,7 @@ def _check_vocabulary(self):
                 "sure to pass a `proto` argument when creating the layer."
             )
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def tokenize(self, inputs):
         self._check_vocabulary()
         unbatched = inputs.shape.rank == 0
@@ -262,7 +262,7 @@ def tokenize(self, inputs):
             tf.ensure_shape(tokens, shape=[self.sequence_length])
         return tokens
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def detokenize(self, inputs):
         self._check_vocabulary()
         inputs, unbatched, rectangular = convert_to_ragged_batch(inputs)
diff --git a/keras_nlp/src/tokenizers/tokenizer.py b/keras_nlp/src/tokenizers/tokenizer.py
index f1304c8413..5a0b3712e4 100644
--- a/keras_nlp/src/tokenizers/tokenizer.py
+++ b/keras_nlp/src/tokenizers/tokenizer.py
@@ -27,7 +27,7 @@
 from keras_nlp.src.utils.preset_utils import save_serialized_object
 from keras_nlp.src.utils.preset_utils import save_tokenizer_assets
 from keras_nlp.src.utils.python_utils import classproperty
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 
 @keras_nlp_export(
@@ -201,7 +201,7 @@ def save_to_preset(self, preset_dir):
         )
         save_tokenizer_assets(self, preset_dir)
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def call(self, inputs, *args, training=None, **kwargs):
         return self.tokenize(inputs, *args, **kwargs)
 
diff --git a/keras_nlp/src/tokenizers/unicode_codepoint_tokenizer.py b/keras_nlp/src/tokenizers/unicode_codepoint_tokenizer.py
index 16115fa199..30bf31afc1 100644
--- a/keras_nlp/src/tokenizers/unicode_codepoint_tokenizer.py
+++ b/keras_nlp/src/tokenizers/unicode_codepoint_tokenizer.py
@@ -17,7 +17,7 @@
 from keras_nlp.src.tokenizers import tokenizer
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
 from keras_nlp.src.utils.tensor_utils import is_int_dtype
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow as tf
@@ -284,7 +284,7 @@ def get_vocabulary(self):
             vocab[chr(i)] = i
         return vocab
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def tokenize(self, inputs):
         unbatched = inputs.shape.rank == 0
         if unbatched:
@@ -321,7 +321,7 @@ def tokenize(self, inputs):
 
         return tokens
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def detokenize(self, inputs):
         inputs, unbatched, rectangular = convert_to_ragged_batch(inputs)
         inputs = tf.ragged.boolean_mask(inputs, tf.not_equal(inputs, 0))
diff --git a/keras_nlp/src/tokenizers/word_piece_tokenizer.py b/keras_nlp/src/tokenizers/word_piece_tokenizer.py
index 8336ffe83b..f228afae2a 100644
--- a/keras_nlp/src/tokenizers/word_piece_tokenizer.py
+++ b/keras_nlp/src/tokenizers/word_piece_tokenizer.py
@@ -23,7 +23,7 @@
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
 from keras_nlp.src.utils.tensor_utils import is_int_dtype
 from keras_nlp.src.utils.tensor_utils import is_string_dtype
-from keras_nlp.src.utils.tensor_utils import tf_preprocessing_function
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
 
 try:
     import tensorflow as tf
@@ -470,7 +470,7 @@ def _check_vocabulary(self):
                 "to pass a `vocabulary` argument when creating the layer."
             )
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def tokenize(self, inputs):
         self._check_vocabulary()
         unbatched = inputs.shape.rank == 0
@@ -515,7 +515,7 @@ def tokenize(self, inputs):
 
         return tokens
 
-    @tf_preprocessing_function
+    @preprocessing_function
     def detokenize(self, inputs):
         self._check_vocabulary()
         inputs, unbatched, rectangular = convert_to_ragged_batch(inputs)
diff --git a/keras_nlp/src/utils/preset_utils.py b/keras_nlp/src/utils/preset_utils.py
index 146f64ce12..5960a1ea02 100644
--- a/keras_nlp/src/utils/preset_utils.py
+++ b/keras_nlp/src/utils/preset_utils.py
@@ -60,6 +60,8 @@
 # Config file names.
 CONFIG_FILE = "config.json"
 TOKENIZER_CONFIG_FILE = "tokenizer.json"
+AUDIO_CONVERTER_CONFIG_FILE = "audio_converter.json"
+IMAGE_CONVERTER_CONFIG_FILE = "image_converter.json"
 TASK_CONFIG_FILE = "task.json"
 PREPROCESSOR_CONFIG_FILE = "preprocessor.json"
 METADATA_FILE = "metadata.json"
@@ -656,6 +658,14 @@ def load_tokenizer(self, cls, **kwargs):
         """Load a tokenizer layer from the preset."""
         raise NotImplementedError
 
+    def load_audio_converter(self, cls, **kwargs):
+        """Load a audio converter layer from the preset."""
+        raise NotImplementedError
+
+    def load_image_converter(self, cls, **kwargs):
+        """Load a image converter layer from the preset."""
+        raise NotImplementedError
+
     def load_task(self, cls, load_weights, load_task_extras, **kwargs):
         """Load a task model from the preset.
 
@@ -685,6 +695,14 @@ def load_preprocessor(self, cls, load_task_extras, **kwargs):
         """
         if "tokenizer" not in kwargs:
             kwargs["tokenizer"] = self.load_tokenizer(cls.tokenizer_cls)
+        if "audio_converter" not in kwargs:
+            kwargs["audio_converter"] = self.load_audio_converter(
+                cls.audio_converter_class
+            )
+        if "image_converter" not in kwargs:
+            kwargs["image_converter"] = self.load_image_converter(
+                cls.image_converter_class
+            )
         return cls(**kwargs)
 
 
@@ -705,6 +723,14 @@ def load_tokenizer(self, cls, **kwargs):
         tokenizer.load_preset_assets(self.preset)
         return tokenizer
 
+    def load_audio_converter(self, cls, **kwargs):
+        converter_config = load_json(self.preset, AUDIO_CONVERTER_CONFIG_FILE)
+        return load_serialized_object(converter_config, **kwargs)
+
+    def load_image_converter(self, cls, **kwargs):
+        converter_config = load_json(self.preset, IMAGE_CONVERTER_CONFIG_FILE)
+        return load_serialized_object(converter_config, **kwargs)
+
     def load_task(self, cls, load_weights, load_task_extras, **kwargs):
         # If there is no `task.json` or it's for the wrong class delegate to the
         # super class loader.
diff --git a/keras_nlp/src/utils/tensor_utils.py b/keras_nlp/src/utils/tensor_utils.py
index 4806400f2f..e21c0520fa 100644
--- a/keras_nlp/src/utils/tensor_utils.py
+++ b/keras_nlp/src/utils/tensor_utils.py
@@ -20,8 +20,6 @@
 import keras
 from keras import ops
 
-from keras_nlp.src.api_export import keras_nlp_export
-
 try:
     import tensorflow as tf
     import tensorflow_text as tf_text
@@ -47,7 +45,7 @@ def in_no_convert_scope():
     return NO_CONVERT_COUNTER.count > 0
 
 
-def tf_preprocessing_function(fn):
+def preprocessing_function(fn):
     """Wraps a preprocessing function to handle tf tensor conversion."""
     if tf is None:
         return fn
@@ -59,33 +57,30 @@ def tf_preprocessing_function(fn):
 
             @functools.wraps(fn)
             def wrapper(self, x, **kwargs):
-                x = convert_to_tf(x)
+                x = convert_inputs(x)
                 with no_convert_scope():
                     x = fn(self, x, **kwargs)
-                return convert_from_tf(x)
+                return convert_outputs(x)
 
         else:
 
             @functools.wraps(fn)
             def wrapper(self, x, y=None, sample_weight=None, **kwargs):
-                x, y, sample_weight = convert_to_tf((x, y, sample_weight))
+                x, y, sample_weight = convert_inputs((x, y, sample_weight))
                 with no_convert_scope():
                     x = fn(self, x, y=y, sample_weight=sample_weight, **kwargs)
-                return convert_from_tf(x)
+                return convert_outputs(x)
 
         return wrapper
 
 
-@keras_nlp_export("keras_nlp.utils.convert_to_tf")
-def convert_to_tf(x):
-    """Convert raw inputs to tf inputs for preprocessing.
+def convert_inputs(x):
+    """Convert raw inputs for preprocessing.
 
     This function is used to convert raw inputs (strings, lists, `np.ndarray`s,
-    `jax.Array`s, `torch.Tensor`s) to tensorflow inputs for use with `tf.data`
-    and KerasNLP preprocessing layers. It will convert ragged inputs and string
-    inputs `tf.RaggedTensor` and `tf.Tensor` types. This will automatically be
-    called when running preprocessing layers or `keras_nlp.models.Task`s with
-    preprocessing included.
+    `jax.Array`s, `torch.Tensor`s, etc) to inputs for preprocessing layers. It
+    will convert ragged inputs and strins inputs `tf.RaggedTensor` and
+    `tf.Tensor` types, and any dense numeric tensors to the backend tensor type.
 
     `tuple` and `list` elements are handled differently by this function. A
     `tuple` is assumed to enumerate separate inputs, and a `list` is assumed to
@@ -97,11 +92,11 @@ def convert_to_tf(x):
     ```python
     # Two ragged arrays of token ids.
     x = ([[1, 2, 3], [4, 5]], [[1, 2], [3, 4, 5]])
-    keras_nlp.utils.convert_to_tf(x)
+    keras_nlp.utils.convert_inputs(x)
 
     # A batch of three samples each with two string segments.
     x = (["hi", "hello", "hey"], ["bye", "later", "so long"])
-    keras_nlp.utils.convert_to_tf(x)
+    keras_nlp.utils.convert_inputs(x)
 
     # A batch of features in a dictionary.
     x = {
@@ -109,17 +104,20 @@ def convert_to_tf(x):
         "images": np.ones((3, 64, 64, 3)),
         "labels": [1, 0, 1],
     }
-    keras_nlp.utils.convert_to_tf(x)
+    keras_nlp.utils.convert_inputs(x)
     ```
     """
     if isinstance(x, dict):
-        return {k: convert_to_tf(x[k]) for k, v in x.items()}
+        return {k: convert_inputs(x[k]) for k, v in x.items()}
     if isinstance(x, tuple):
-        return tuple(convert_to_tf(v) for v in x)
-    if isinstance(x, list):
-        return tf.ragged.constant(x)
+        return tuple(convert_inputs(v) for v in x)
     if isinstance(x, str):
         return tf.constant(x)
+    if isinstance(x, list):
+        try:
+            return ops.convert_to_tensor(x)
+        except:
+            return tf.ragged.constant(x)
     if is_tensor_type(x):
         # Torch will complain about device placement for GPU tensors.
         if keras.config.backend() == "torch":
@@ -127,13 +125,12 @@ def convert_to_tf(x):
 
             if isinstance(x, torch.Tensor):
                 x = x.cpu()
-        return tf.convert_to_tensor(x)
+        return x
     return x
 
 
-@keras_nlp_export("keras_nlp.utils.convert_from_tf")
-def convert_from_tf(x):
-    """Convert tf outputs after preprocessing to a backend agnostic format.
+def convert_outputs(x):
+    """Convert outputs after preprocessing to a backend agnostic format.
 
     This function is used to convert `tf.Tensor` and `tf.RaggedTensor` output
     from preprocessing layers to either:
@@ -150,11 +147,11 @@ def convert_from_tf(x):
     ```python
     # Two ragged arrays of token ids.
     x = tf.ragged.constant([[1, 2, 3], [4, 5]])
-    keras_nlp.utils.convert_from_tf(x)
+    keras_nlp.utils.convert_outputs(x)
 
     # A batch of three samples each with two string segments.
     x = (tf.constant["hi", "yo", "hey"]), tf.constant(["bye", "ciao", ""]))
-    keras_nlp.utils.convert_from_tf(x)
+    keras_nlp.utils.convert_outputs(x)
 
     # A batch of features in a dictionary.
     x = {
@@ -162,7 +159,7 @@ def convert_from_tf(x):
         "images": tf.ones((3, 64, 64, 3)),
         "labels": tf.constant([1, 0, 1]),
     }
-    keras_nlp.utils.convert_from_tf(x)
+    keras_nlp.utils.convert_outputs(x)
     ```
     """
     if not tf.executing_eagerly() or in_no_convert_scope():
diff --git a/keras_nlp/src/utils/tensor_utils_test.py b/keras_nlp/src/utils/tensor_utils_test.py
index 478f05c9ad..c8139427fc 100644
--- a/keras_nlp/src/utils/tensor_utils_test.py
+++ b/keras_nlp/src/utils/tensor_utils_test.py
@@ -19,45 +19,44 @@
 
 from keras_nlp.src.tests.test_case import TestCase
 from keras_nlp.src.utils.tensor_utils import any_equal
-from keras_nlp.src.utils.tensor_utils import convert_from_tf
+from keras_nlp.src.utils.tensor_utils import convert_outputs
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
-from keras_nlp.src.utils.tensor_utils import convert_to_tf
+from keras_nlp.src.utils.tensor_utils import convert_inputs
 from keras_nlp.src.utils.tensor_utils import is_tensor_type
 from keras_nlp.src.utils.tensor_utils import tensor_to_list
 
 
-class ConvertTf(TestCase):
+class ConvertHelpers(TestCase):
     def test_basics(self):
         inputs = ops.array([1, 2, 3])
         # Convert to tf.
-        outputs = convert_to_tf(inputs)
-        self.assertIsInstance(outputs, tf.Tensor)
-        self.assertAllEqual(outputs, tf.constant(ops.convert_to_numpy(inputs)))
+        outputs = convert_inputs(inputs)
+        self.assertAllEqual(outputs, ops.array(inputs))
         # Convert from tf.
-        outputs = convert_from_tf(outputs)
+        outputs = convert_outputs(outputs)
         self.assertTrue(is_tensor_type(outputs))
         self.assertAllEqual(outputs, inputs)
 
     def test_strings(self):
         inputs = ["one", "two"]
         # Convert to tf.
-        outputs = convert_to_tf(inputs)
+        outputs = convert_inputs(inputs)
         self.assertIsInstance(outputs, tf.Tensor)
         self.assertAllEqual(outputs, tf.constant(inputs))
         # Convert from tf.
-        outputs = convert_from_tf(outputs)
+        outputs = convert_outputs(outputs)
         self.assertIsInstance(outputs, list)
         self.assertEqual(outputs, inputs)
 
     def test_ragged(self):
         inputs = [np.ones((1, 3)), np.ones((1, 2))]
         # Convert to tf.
-        outputs = convert_to_tf(inputs)
+        outputs = convert_inputs(inputs)
         self.assertIsInstance(outputs, tf.RaggedTensor)
         print(outputs, inputs)
         self.assertAllEqual(outputs, tf.ragged.constant(inputs))
         # Convert from tf.
-        outputs = convert_from_tf(outputs)
+        outputs = convert_outputs(outputs)
         self.assertIsInstance(outputs, list)
         self.assertEqual(outputs, [[[1, 1, 1]], [[1, 1]]])
 
@@ -72,14 +71,14 @@ def test_composite(self):
             [3, 4],
         )
 
-        outputs = convert_to_tf(inputs)
+        outputs = convert_inputs(inputs)
         self.assertIsInstance(outputs[0]["text"], tf.Tensor)
         self.assertIsInstance(outputs[0]["images"], tf.RaggedTensor)
         self.assertIsInstance(outputs[0]["ragged_ints"], tf.RaggedTensor)
-        self.assertIsInstance(outputs[1], tf.Tensor)
-        self.assertIsInstance(outputs[2], tf.Tensor)
+        self.assertTrue(is_tensor_type(outputs[1]))
+        self.assertTrue(is_tensor_type(outputs[2]))
 
-        outputs = convert_from_tf(outputs)
+        outputs = convert_outputs(outputs)
         self.assertIsInstance(outputs[0]["text"], list)
         self.assertIsInstance(outputs[0]["images"], list)
         self.assertIsInstance(outputs[0]["ragged_ints"], list)

From f73defc88bf95450efe02f14d31469ac5e9d0a7f Mon Sep 17 00:00:00 2001
From: Matt Watson <mattdangerw@gmail.com>
Date: Thu, 5 Sep 2024 11:43:27 -0700
Subject: [PATCH 2/8] Fix some tests

---
 keras_nlp/api/layers/__init__.py              |   3 +
 .../layers/preprocessing/audio_converter.py   |  13 ++
 .../preprocessing/audio_converter_test.py     |  69 +++++++++++
 .../layers/preprocessing/image_converter.py   |  44 ++-----
 .../preprocessing/image_converter_test.py     |  72 +++++++++++
 .../preprocessing/resizing_image_converter.py | 116 ++++++++++++++++++
 .../resizing_image_converter_test.py          |  47 +++++++
 .../preprocessing/start_end_packer_test.py    |   4 -
 .../src/models/causal_lm_preprocessor.py      |   2 +-
 .../pali_gemma/pali_gemma_backbone_test.py    |   2 +-
 .../pali_gemma/pali_gemma_causal_lm_test.py   |   6 +-
 .../pali_gemma/pali_gemma_image_converter.py  |   6 +-
 keras_nlp/src/models/preprocessor.py          |   2 +
 .../src/models/seq_2_seq_lm_preprocessor.py   |   2 +-
 .../tokenizers/sentence_piece_tokenizer.py    |   2 +-
 keras_nlp/src/utils/preset_utils.py           |  10 +-
 keras_nlp/src/utils/tensor_utils.py           |   2 +
 keras_nlp/src/utils/tensor_utils_test.py      |   2 +-
 18 files changed, 353 insertions(+), 51 deletions(-)
 create mode 100644 keras_nlp/src/layers/preprocessing/audio_converter_test.py
 create mode 100644 keras_nlp/src/layers/preprocessing/image_converter_test.py
 create mode 100644 keras_nlp/src/layers/preprocessing/resizing_image_converter.py
 create mode 100644 keras_nlp/src/layers/preprocessing/resizing_image_converter_test.py

diff --git a/keras_nlp/api/layers/__init__.py b/keras_nlp/api/layers/__init__.py
index 5cffa3fb40..6431128b4e 100644
--- a/keras_nlp/api/layers/__init__.py
+++ b/keras_nlp/api/layers/__init__.py
@@ -46,6 +46,9 @@
 )
 from keras_nlp.src.layers.preprocessing.random_deletion import RandomDeletion
 from keras_nlp.src.layers.preprocessing.random_swap import RandomSwap
+from keras_nlp.src.layers.preprocessing.resizing_image_converter import (
+    ResizingImageConverter,
+)
 from keras_nlp.src.layers.preprocessing.start_end_packer import StartEndPacker
 from keras_nlp.src.models.pali_gemma.pali_gemma_image_converter import (
     PaliGemmaImageConverter,
diff --git a/keras_nlp/src/layers/preprocessing/audio_converter.py b/keras_nlp/src/layers/preprocessing/audio_converter.py
index 9569c44801..efc8a83117 100644
--- a/keras_nlp/src/layers/preprocessing/audio_converter.py
+++ b/keras_nlp/src/layers/preprocessing/audio_converter.py
@@ -18,7 +18,10 @@
 from keras_nlp.src.utils.preset_utils import AUDIO_CONVERTER_CONFIG_FILE
 from keras_nlp.src.utils.preset_utils import find_subclass
 from keras_nlp.src.utils.preset_utils import get_preset_loader
+from keras_nlp.src.utils.preset_utils import list_presets
+from keras_nlp.src.utils.preset_utils import list_subclasses
 from keras_nlp.src.utils.preset_utils import save_serialized_object
+from keras_nlp.src.utils.python_utils import classproperty
 
 
 @keras_nlp_export("keras_nlp.layers.AudioConverter")
@@ -45,6 +48,16 @@ class AudioConverter(PreprocessingLayer):
     ```
     """
 
+    backbone_cls = None
+
+    @classproperty
+    def presets(cls):
+        """List built-in presets for a `Task` subclass."""
+        presets = list_presets(cls)
+        for subclass in list_subclasses(cls):
+            presets.update(subclass.presets)
+        return presets
+
     @classmethod
     def from_preset(
         cls,
diff --git a/keras_nlp/src/layers/preprocessing/audio_converter_test.py b/keras_nlp/src/layers/preprocessing/audio_converter_test.py
new file mode 100644
index 0000000000..7e271e757f
--- /dev/null
+++ b/keras_nlp/src/layers/preprocessing/audio_converter_test.py
@@ -0,0 +1,69 @@
+# Copyright 2024 The KerasNLP Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import pathlib
+
+import numpy as np
+import pytest
+
+from keras_nlp.src.layers.preprocessing.audio_converter import AudioConverter
+from keras_nlp.src.models.backbone import Backbone
+from keras_nlp.src.models.whisper.whisper_audio_converter import (
+    WhisperAudioConverter,
+)
+from keras_nlp.src.tests.test_case import TestCase
+
+
+class AudioConverterTest(TestCase):
+    def test_preset_accessors(self):
+        pali_gemma_presets = set(WhisperAudioConverter.presets.keys())
+        all_presets = set(AudioConverter.presets.keys())
+        self.assertContainsSubset(pali_gemma_presets, all_presets)
+
+    @pytest.mark.large
+    def test_from_preset(self):
+        self.assertIsInstance(
+            AudioConverter.from_preset("whisper_tiny_en"),
+            WhisperAudioConverter,
+        )
+
+    @pytest.mark.large
+    def test_from_preset_errors(self):
+        with self.assertRaises(ValueError):
+            AudioConverter.from_preset("bert_tiny_en_uncased")
+        with self.assertRaises(ValueError):
+            # No loading on a non-keras model.
+            AudioConverter.from_preset("hf://spacy/en_core_web_sm")
+
+    @pytest.mark.large
+    def test_save_to_preset(self):
+        save_dir = self.get_temp_dir()
+        converter = AudioConverter.from_preset(
+            "whisper_tiny_en",
+            num_mels=40,
+        )
+        converter.save_to_preset(save_dir)
+        # Save a backbone so the preset is valid.
+        backbone = Backbone.from_preset("whisper_tiny_en", load_weights=False)
+        backbone.save_to_preset(save_dir)
+
+        # Check existence of files.
+        path = pathlib.Path(save_dir)
+        self.assertTrue(os.path.exists(path / "audio_converter.json"))
+
+        # Check loading.
+        restored = AudioConverter.from_preset(save_dir)
+        test_image = np.random.rand(100, 100, 3) * 255
+        self.assertAllClose(restored(test_image), converter(test_image))
diff --git a/keras_nlp/src/layers/preprocessing/image_converter.py b/keras_nlp/src/layers/preprocessing/image_converter.py
index 58ae3fb088..1807d3c492 100644
--- a/keras_nlp/src/layers/preprocessing/image_converter.py
+++ b/keras_nlp/src/layers/preprocessing/image_converter.py
@@ -11,8 +11,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-import keras
-
 from keras_nlp.src.api_export import keras_nlp_export
 from keras_nlp.src.layers.preprocessing.preprocessing_layer import (
     PreprocessingLayer,
@@ -20,8 +18,10 @@
 from keras_nlp.src.utils.preset_utils import IMAGE_CONVERTER_CONFIG_FILE
 from keras_nlp.src.utils.preset_utils import find_subclass
 from keras_nlp.src.utils.preset_utils import get_preset_loader
+from keras_nlp.src.utils.preset_utils import list_presets
+from keras_nlp.src.utils.preset_utils import list_subclasses
 from keras_nlp.src.utils.preset_utils import save_serialized_object
-from keras_nlp.src.utils.tensor_utils import preprocessing_function
+from keras_nlp.src.utils.python_utils import classproperty
 
 
 @keras_nlp_export("keras_nlp.layers.ImageConverter")
@@ -45,41 +45,21 @@ class ImageConverter(PreprocessingLayer):
     # Resize images for `"pali_gemma_3b_224"`.
     converter = keras_nlp.layers.ImageConverter.from_preset("pali_gemma_3b_224")
     converter(np.ones(2, 512, 512, 3)) # Output shape: (2, 224, 224, 3)
-    # Resize images for `"pali_gemma_3b_224"`.
+    # Resize images for `"pali_gemma_3b_448"`.
     converter = keras_nlp.layers.ImageConverter.from_preset("pali_gemma_3b_448")
     converter(np.ones(2, 512, 512, 3)) # Output shape: (2, 448, 448, 3)
     ```
     """
 
-    def __init__(
-        self,
-        height,
-        width,
-        crop_to_aspect_ratio=True,
-        pad_to_aspect_ratio=False,
-        interpolation="bilinear",
-        fill_mode="constant",
-        fill_value=0.0,
-        data_format=None,
-        **kwargs,
-    ):
-        super().__init__(**kwargs)
-        # By default, we just do a simple resize. Any model can subclass this
-        # layer for preprocessing of a raw image to a model image input.
-        self.resizing = keras.layers.Resizing(
-            height,
-            width,
-            crop_to_aspect_ratio=crop_to_aspect_ratio,
-            pad_to_aspect_ratio=pad_to_aspect_ratio,
-            interpolation=interpolation,
-            fill_mode=fill_mode,
-            fill_value=fill_value,
-            data_format=data_format,
-        )
+    backbone_cls = None
 
-    @preprocessing_function
-    def call(self, inputs):
-        return self.resizing(inputs)
+    @classproperty
+    def presets(cls):
+        """List built-in presets for a `Task` subclass."""
+        presets = list_presets(cls)
+        for subclass in list_subclasses(cls):
+            presets.update(subclass.presets)
+        return presets
 
     @classmethod
     def from_preset(
diff --git a/keras_nlp/src/layers/preprocessing/image_converter_test.py b/keras_nlp/src/layers/preprocessing/image_converter_test.py
new file mode 100644
index 0000000000..d5b25f5840
--- /dev/null
+++ b/keras_nlp/src/layers/preprocessing/image_converter_test.py
@@ -0,0 +1,72 @@
+# Copyright 2024 The KerasNLP Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import pathlib
+
+import numpy as np
+import pytest
+
+from keras_nlp.src.layers.preprocessing.image_converter import ImageConverter
+from keras_nlp.src.models.backbone import Backbone
+from keras_nlp.src.models.pali_gemma.pali_gemma_image_converter import (
+    PaliGemmaImageConverter,
+)
+from keras_nlp.src.tests.test_case import TestCase
+
+
+class ImageConverterTest(TestCase):
+    def test_preset_accessors(self):
+        pali_gemma_presets = set(PaliGemmaImageConverter.presets.keys())
+        all_presets = set(ImageConverter.presets.keys())
+        self.assertContainsSubset(pali_gemma_presets, all_presets)
+
+    @pytest.mark.large
+    def test_from_preset(self):
+        self.assertIsInstance(
+            ImageConverter.from_preset("pali_gemma_3b_mix_224"),
+            PaliGemmaImageConverter,
+        )
+
+    @pytest.mark.large
+    def test_from_preset_errors(self):
+        with self.assertRaises(ValueError):
+            ImageConverter.from_preset("bert_tiny_en_uncased")
+        with self.assertRaises(ValueError):
+            # No loading on a non-keras model.
+            ImageConverter.from_preset("hf://spacy/en_core_web_sm")
+
+    @pytest.mark.large
+    def test_save_to_preset(self):
+        save_dir = self.get_temp_dir()
+        converter = ImageConverter.from_preset(
+            "pali_gemma_3b_mix_224",
+            interpolation="nearest",
+        )
+        converter.save_to_preset(save_dir)
+        # Save a backbone so the preset is valid.
+        backbone = Backbone.from_preset(
+            "pali_gemma_3b_mix_224",
+            load_weights=False,
+        )
+        backbone.save_to_preset(save_dir)
+
+        # Check existence of files.
+        path = pathlib.Path(save_dir)
+        self.assertTrue(os.path.exists(path / "image_converter.json"))
+
+        # Check loading.
+        restored = ImageConverter.from_preset(save_dir)
+        test_image = np.random.rand(100, 100, 3) * 255
+        self.assertAllClose(restored(test_image), converter(test_image))
diff --git a/keras_nlp/src/layers/preprocessing/resizing_image_converter.py b/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
new file mode 100644
index 0000000000..be839b3417
--- /dev/null
+++ b/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
@@ -0,0 +1,116 @@
+# Copyright 2024 The KerasNLP Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import keras
+
+from keras_nlp.src.api_export import keras_nlp_export
+from keras_nlp.src.layers.preprocessing.image_converter import ImageConverter
+from keras_nlp.src.utils.tensor_utils import preprocessing_function
+
+
+@keras_nlp_export("keras_nlp.layers.ResizingImageConverter")
+class ResizingImageConverter(ImageConverter):
+    """An `ImageConverter` that simply resizes the input image.
+
+    The `ResizingImageConverter` is a subclass of `ImageConverter` for models
+    that simply need to resize image tensors before using them for modeling.
+    The layer will take a input a raw image tensor (batched or unbatched) in the
+    channels last or channels first format, and output a resize tensor.
+
+    Args:
+        height: Integer, the height of the output shape.
+        width: Integer, the width of the output shape.
+        crop_to_aspect_ratio: If `True`, resize the images without aspect
+            ratio distortion. When the original aspect ratio differs
+            from the target aspect ratio, the output image will be
+            cropped so as to return the
+            largest possible window in the image (of size `(height, width)`)
+            that matches the target aspect ratio. By default
+            (`crop_to_aspect_ratio=False`), aspect ratio may not be preserved.
+        pad_to_aspect_ratio: If `True`, pad the images without aspect
+            ratio distortion. When the original aspect ratio differs
+            from the target aspect ratio, the output image will be
+            evenly padded on the short side.
+        interpolation: String, the interpolation method.
+            Supports `"bilinear"`, `"nearest"`, `"bicubic"`,
+            `"lanczos3"`, `"lanczos5"`. Defaults to `"bilinear"`.
+        fill_mode: When using `pad_to_aspect_ratio=True`, padded areas
+            are filled according to the given mode. Only `"constant"` is
+            supported at this time
+            (fill with constant value, equal to `fill_value`).
+        fill_value: Float. Padding value to use when `pad_to_aspect_ratio=True`.
+        data_format: string, either `"channels_last"` or `"channels_first"`.
+            The ordering of the dimensions in the inputs. `"channels_last"`
+            corresponds to inputs with shape `(batch, height, width, channels)`
+            while `"channels_first"` corresponds to inputs with shape
+            `(batch, channels, height, width)`. It defaults to the
+            `image_data_format` value found in your Keras config file at
+            `~/.keras/keras.json`. If you never set it, then it will be
+            `"channels_last"`.
+
+    Examples:
+    ```python
+    # Resize images for `"pali_gemma_3b_224"`.
+    converter = keras_nlp.layers.ImageConverter.from_preset("pali_gemma_3b_224")
+    converter(np.ones(2, 512, 512, 3)) # Output shape: (2, 224, 224, 3)
+    # Resize images for `"pali_gemma_3b_224"`.
+    converter = keras_nlp.layers.ImageConverter.from_preset("pali_gemma_3b_448")
+    converter(np.ones(2, 512, 512, 3)) # Output shape: (2, 448, 448, 3)
+    ```
+    """
+
+    def __init__(
+        self,
+        height,
+        width,
+        crop_to_aspect_ratio=True,
+        pad_to_aspect_ratio=False,
+        interpolation="bilinear",
+        fill_mode="constant",
+        fill_value=0.0,
+        data_format=None,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        # By default, we just do a simple resize. Any model can subclass this
+        # layer for preprocessing of a raw image to a model image input.
+        self.resizing = keras.layers.Resizing(
+            height,
+            width,
+            crop_to_aspect_ratio=crop_to_aspect_ratio,
+            pad_to_aspect_ratio=pad_to_aspect_ratio,
+            interpolation=interpolation,
+            fill_mode=fill_mode,
+            fill_value=fill_value,
+            data_format=data_format,
+        )
+
+    @preprocessing_function
+    def call(self, inputs):
+        return self.resizing(inputs)
+
+    def get_config(self):
+        config = super().get_config()
+        config.update(
+            {
+                "height": self.resizing.height,
+                "width": self.resizing.width,
+                "interpolation": self.resizing.interpolation,
+                "crop_to_aspect_ratio": self.resizing.crop_to_aspect_ratio,
+                "pad_to_aspect_ratio": self.resizing.pad_to_aspect_ratio,
+                "fill_mode": self.resizing.fill_mode,
+                "fill_value": self.resizing.fill_value,
+                "data_format": self.resizing.data_format,
+            }
+        )
+        return config
diff --git a/keras_nlp/src/layers/preprocessing/resizing_image_converter_test.py b/keras_nlp/src/layers/preprocessing/resizing_image_converter_test.py
new file mode 100644
index 0000000000..9a4b8c0c02
--- /dev/null
+++ b/keras_nlp/src/layers/preprocessing/resizing_image_converter_test.py
@@ -0,0 +1,47 @@
+# Copyright 2024 The KerasNLP Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+from keras import ops
+
+from keras_nlp.src.layers.preprocessing.resizing_image_converter import (
+    ResizingImageConverter,
+)
+from keras_nlp.src.tests.test_case import TestCase
+
+
+class ResizingImageConverterTest(TestCase):
+    def test_resize_one(self):
+        converter = ResizingImageConverter(22, 22)
+        test_image = np.random.rand(10, 10, 3) * 255
+        shape = ops.shape(converter(test_image))
+        self.assertEqual(shape, (22, 22, 3))
+
+    def test_resize_batch(self):
+        converter = ResizingImageConverter(12, 12)
+        test_batch = np.random.rand(4, 10, 20, 3) * 255
+        shape = ops.shape(converter(test_batch))
+        self.assertEqual(shape, (4, 12, 12, 3))
+
+    def test_config(self):
+        converter = ResizingImageConverter(
+            width=12,
+            height=20,
+            pad_to_aspect_ratio=True,
+            crop_to_aspect_ratio=False,
+            fill_value=7.0,
+        )
+        clone = ResizingImageConverter.from_config(converter.get_config())
+        test_batch = np.random.rand(4, 10, 20, 3) * 255
+        self.assertAllClose(converter(test_batch), clone(test_batch))
diff --git a/keras_nlp/src/layers/preprocessing/start_end_packer_test.py b/keras_nlp/src/layers/preprocessing/start_end_packer_test.py
index dc99eaa8cd..5fb77a930e 100644
--- a/keras_nlp/src/layers/preprocessing/start_end_packer_test.py
+++ b/keras_nlp/src/layers/preprocessing/start_end_packer_test.py
@@ -12,7 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import keras
 import tensorflow as tf
 
 from keras_nlp.src.layers.preprocessing.start_end_packer import StartEndPacker
@@ -30,13 +29,10 @@ def test_dense_input(self):
     def test_bfloat16_dtype(self):
         # Core Keras has a strange bug where it converts int to floats in
         # ops.convert_to_tensor only with jax and bfloat16.
-        floatx = keras.config.floatx()
-        keras.config.set_floatx("bfloat16")
         input_data = [5, 6, 7]
         start_end_packer = StartEndPacker(sequence_length=5, dtype="bfloat16")
         output = start_end_packer(input_data)
         self.assertDTypeEqual(output, "int32")
-        keras.config.set_floatx(floatx)
 
     def test_dense_2D_input(self):
         input_data = [[5, 6, 7]]
diff --git a/keras_nlp/src/models/causal_lm_preprocessor.py b/keras_nlp/src/models/causal_lm_preprocessor.py
index 817851493a..6a0dad3bdf 100644
--- a/keras_nlp/src/models/causal_lm_preprocessor.py
+++ b/keras_nlp/src/models/causal_lm_preprocessor.py
@@ -16,8 +16,8 @@
 from keras_nlp.src.api_export import keras_nlp_export
 from keras_nlp.src.layers.preprocessing.start_end_packer import StartEndPacker
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import strip_to_ragged
 from keras_nlp.src.utils.tensor_utils import preprocessing_function
+from keras_nlp.src.utils.tensor_utils import strip_to_ragged
 
 
 @keras_nlp_export("keras_nlp.models.CausalLMPreprocessor")
diff --git a/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py b/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py
index 4a899c6734..25c4b27a6f 100644
--- a/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py
@@ -99,7 +99,7 @@ def test_smallest_preset(self):
             expected_output_shape=(1, 261, 2048),
             # The forward pass from a preset should be stable!
             expected_partial_output=ops.array(
-                [-0.449851,  1.431027, -0.713446,  0.417485, -0.640859]
+                [-0.449851, 1.431027, -0.713446, 0.417485, -0.640859]
             ),
         )
 
diff --git a/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_test.py b/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_test.py
index 04d0745a8f..3009cbf944 100644
--- a/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_test.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_test.py
@@ -25,12 +25,12 @@
 from keras_nlp.src.models.pali_gemma.pali_gemma_causal_lm_preprocessor import (
     PaliGemmaCausalLMPreprocessor,
 )
-from keras_nlp.src.models.pali_gemma.pali_gemma_tokenizer import (
-    PaliGemmaTokenizer,
-)
 from keras_nlp.src.models.pali_gemma.pali_gemma_image_converter import (
     PaliGemmaImageConverter,
 )
+from keras_nlp.src.models.pali_gemma.pali_gemma_tokenizer import (
+    PaliGemmaTokenizer,
+)
 from keras_nlp.src.tests.test_case import TestCase
 
 
diff --git a/keras_nlp/src/models/pali_gemma/pali_gemma_image_converter.py b/keras_nlp/src/models/pali_gemma/pali_gemma_image_converter.py
index 21d95c9fa6..6bef25d8b4 100644
--- a/keras_nlp/src/models/pali_gemma/pali_gemma_image_converter.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_image_converter.py
@@ -12,12 +12,14 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from keras_nlp.src.api_export import keras_nlp_export
-from keras_nlp.src.layers.preprocessing.image_converter import ImageConverter
+from keras_nlp.src.layers.preprocessing.resizing_image_converter import (
+    ResizingImageConverter,
+)
 from keras_nlp.src.models.pali_gemma.pali_gemma_backbone import (
     PaliGemmaBackbone,
 )
 
 
 @keras_nlp_export("keras_nlp.layers.PaliGemmaImageConverter")
-class PaliGemmaImageConverter(ImageConverter):
+class PaliGemmaImageConverter(ResizingImageConverter):
     backbone_cls = PaliGemmaBackbone
diff --git a/keras_nlp/src/models/preprocessor.py b/keras_nlp/src/models/preprocessor.py
index ef1ce6d503..f1fb32aa72 100644
--- a/keras_nlp/src/models/preprocessor.py
+++ b/keras_nlp/src/models/preprocessor.py
@@ -44,6 +44,8 @@ class Preprocessor(PreprocessingLayer):
 
     backbone_cls = None
     tokenizer_cls = None
+    audio_converter_cls = None
+    image_converter_cls = None
 
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
diff --git a/keras_nlp/src/models/seq_2_seq_lm_preprocessor.py b/keras_nlp/src/models/seq_2_seq_lm_preprocessor.py
index 84f49f3b90..27405f99d0 100644
--- a/keras_nlp/src/models/seq_2_seq_lm_preprocessor.py
+++ b/keras_nlp/src/models/seq_2_seq_lm_preprocessor.py
@@ -16,8 +16,8 @@
 from keras_nlp.src.api_export import keras_nlp_export
 from keras_nlp.src.layers.preprocessing.start_end_packer import StartEndPacker
 from keras_nlp.src.models.preprocessor import Preprocessor
-from keras_nlp.src.utils.tensor_utils import strip_to_ragged
 from keras_nlp.src.utils.tensor_utils import preprocessing_function
+from keras_nlp.src.utils.tensor_utils import strip_to_ragged
 
 try:
     import tensorflow as tf
diff --git a/keras_nlp/src/tokenizers/sentence_piece_tokenizer.py b/keras_nlp/src/tokenizers/sentence_piece_tokenizer.py
index 31ed3401ca..ea9f4d6f4a 100644
--- a/keras_nlp/src/tokenizers/sentence_piece_tokenizer.py
+++ b/keras_nlp/src/tokenizers/sentence_piece_tokenizer.py
@@ -31,8 +31,8 @@
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
 from keras_nlp.src.utils.tensor_utils import is_int_dtype
 from keras_nlp.src.utils.tensor_utils import is_string_dtype
-from keras_nlp.src.utils.tensor_utils import tensor_to_list
 from keras_nlp.src.utils.tensor_utils import preprocessing_function
+from keras_nlp.src.utils.tensor_utils import tensor_to_list
 
 try:
     import tensorflow_text as tf_text
diff --git a/keras_nlp/src/utils/preset_utils.py b/keras_nlp/src/utils/preset_utils.py
index 5960a1ea02..c4160ad0ae 100644
--- a/keras_nlp/src/utils/preset_utils.py
+++ b/keras_nlp/src/utils/preset_utils.py
@@ -693,15 +693,15 @@ def load_preprocessor(self, cls, load_task_extras, **kwargs):
         arguments. This allow us to support transformers checkpoints by
         only converting the backbone and tokenizer.
         """
-        if "tokenizer" not in kwargs:
+        if "tokenizer" not in kwargs and cls.tokenizer_cls:
             kwargs["tokenizer"] = self.load_tokenizer(cls.tokenizer_cls)
-        if "audio_converter" not in kwargs:
+        if "audio_converter" not in kwargs and cls.audio_converter_cls:
             kwargs["audio_converter"] = self.load_audio_converter(
-                cls.audio_converter_class
+                cls.audio_converter_cls
             )
-        if "image_converter" not in kwargs:
+        if "image_converter" not in kwargs and cls.image_converter_cls:
             kwargs["image_converter"] = self.load_image_converter(
-                cls.image_converter_class
+                cls.image_converter_cls
             )
         return cls(**kwargs)
 
diff --git a/keras_nlp/src/utils/tensor_utils.py b/keras_nlp/src/utils/tensor_utils.py
index e21c0520fa..28e8280c4f 100644
--- a/keras_nlp/src/utils/tensor_utils.py
+++ b/keras_nlp/src/utils/tensor_utils.py
@@ -207,6 +207,8 @@ def tensor_to_list(inputs):
 
 def convert_to_ragged_batch(inputs):
     """Ensure a tf.Tensor is a ragged rank 2 tensor."""
+    if not isinstance(inputs, (tf.RaggedTensor, tf.Tensor)):
+        inputs = tf.convert_to_tensor(inputs)
     unbatched = inputs.shape.rank == 1
     rectangular = isinstance(inputs, tf.Tensor)
     if unbatched:
diff --git a/keras_nlp/src/utils/tensor_utils_test.py b/keras_nlp/src/utils/tensor_utils_test.py
index c8139427fc..1eea9ff0da 100644
--- a/keras_nlp/src/utils/tensor_utils_test.py
+++ b/keras_nlp/src/utils/tensor_utils_test.py
@@ -19,9 +19,9 @@
 
 from keras_nlp.src.tests.test_case import TestCase
 from keras_nlp.src.utils.tensor_utils import any_equal
+from keras_nlp.src.utils.tensor_utils import convert_inputs
 from keras_nlp.src.utils.tensor_utils import convert_outputs
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
-from keras_nlp.src.utils.tensor_utils import convert_inputs
 from keras_nlp.src.utils.tensor_utils import is_tensor_type
 from keras_nlp.src.utils.tensor_utils import tensor_to_list
 

From bba5da05b55a51150a5e2cb623d08b952a65f30c Mon Sep 17 00:00:00 2001
From: Matt Watson <mattdangerw@gmail.com>
Date: Thu, 5 Sep 2024 15:08:19 -0700
Subject: [PATCH 3/8] Input conversion fixes

---
 .../multi_segment_packer_test.py              |  2 +-
 keras_nlp/src/utils/tensor_utils.py           | 65 +++++++++++++------
 keras_nlp/src/utils/tensor_utils_test.py      | 20 +++---
 3 files changed, 55 insertions(+), 32 deletions(-)

diff --git a/keras_nlp/src/layers/preprocessing/multi_segment_packer_test.py b/keras_nlp/src/layers/preprocessing/multi_segment_packer_test.py
index 463a43b412..2c34dcd0bd 100644
--- a/keras_nlp/src/layers/preprocessing/multi_segment_packer_test.py
+++ b/keras_nlp/src/layers/preprocessing/multi_segment_packer_test.py
@@ -31,7 +31,7 @@ def test_trim_single_input_ints(self):
         self.assertAllEqual(segment_ids, [0, 0, 0, 0, 0, 0, 0, 0])
 
     def test_trim_single_input_strings(self):
-        input_data = np.array(["a", "b", "c", "d"])
+        input_data = ["a", "b", "c", "d"]
         packer = MultiSegmentPacker(
             sequence_length=5, start_value="[CLS]", end_value="[SEP]"
         )
diff --git a/keras_nlp/src/utils/tensor_utils.py b/keras_nlp/src/utils/tensor_utils.py
index 28e8280c4f..1fd2a85e37 100644
--- a/keras_nlp/src/utils/tensor_utils.py
+++ b/keras_nlp/src/utils/tensor_utils.py
@@ -18,6 +18,7 @@
 import threading
 
 import keras
+import numpy as np
 from keras import ops
 
 try:
@@ -57,30 +58,33 @@ def preprocessing_function(fn):
 
             @functools.wraps(fn)
             def wrapper(self, x, **kwargs):
-                x = convert_inputs(x)
+                x = convert_preprocessing_inputs(x)
                 with no_convert_scope():
                     x = fn(self, x, **kwargs)
-                return convert_outputs(x)
+                return convert_preprocessing_outputs(x)
 
         else:
 
             @functools.wraps(fn)
             def wrapper(self, x, y=None, sample_weight=None, **kwargs):
-                x, y, sample_weight = convert_inputs((x, y, sample_weight))
+                x, y, sample_weight = convert_preprocessing_inputs(
+                    (x, y, sample_weight)
+                )
                 with no_convert_scope():
                     x = fn(self, x, y=y, sample_weight=sample_weight, **kwargs)
-                return convert_outputs(x)
+                return convert_preprocessing_outputs(x)
 
         return wrapper
 
 
-def convert_inputs(x):
+def convert_preprocessing_inputs(x):
     """Convert raw inputs for preprocessing.
 
     This function is used to convert raw inputs (strings, lists, `np.ndarray`s,
-    `jax.Array`s, `torch.Tensor`s, etc) to inputs for preprocessing layers. It
-    will convert ragged inputs and strins inputs `tf.RaggedTensor` and
-    `tf.Tensor` types, and any dense numeric tensors to the backend tensor type.
+    `jax.Array`s, `torch.Tensor`s, etc) to a canonical format for
+    preprocessing layers. All inputs will be converted to backend tensors if
+    possible, except ragged inputs and string inputs which be converted to tf
+    tensors regardless of backend.
 
     `tuple` and `list` elements are handled differently by this function. A
     `tuple` is assumed to enumerate separate inputs, and a `list` is assumed to
@@ -92,11 +96,11 @@ def convert_inputs(x):
     ```python
     # Two ragged arrays of token ids.
     x = ([[1, 2, 3], [4, 5]], [[1, 2], [3, 4, 5]])
-    keras_nlp.utils.convert_inputs(x)
+    keras_nlp.utils.convert_preprocessing_inputs(x)
 
     # A batch of three samples each with two string segments.
     x = (["hi", "hello", "hey"], ["bye", "later", "so long"])
-    keras_nlp.utils.convert_inputs(x)
+    keras_nlp.utils.convert_preprocessing_inputs(x)
 
     # A batch of features in a dictionary.
     x = {
@@ -104,32 +108,51 @@ def convert_inputs(x):
         "images": np.ones((3, 64, 64, 3)),
         "labels": [1, 0, 1],
     }
-    keras_nlp.utils.convert_inputs(x)
+    keras_nlp.utils.convert_preprocessing_inputs(x)
     ```
     """
+    if not tf.executing_eagerly() or in_no_convert_scope():
+        return x
+
     if isinstance(x, dict):
-        return {k: convert_inputs(x[k]) for k, v in x.items()}
+        return {k: convert_preprocessing_inputs(x[k]) for k, v in x.items()}
     if isinstance(x, tuple):
-        return tuple(convert_inputs(v) for v in x)
+        return tuple(convert_preprocessing_inputs(v) for v in x)
     if isinstance(x, str):
         return tf.constant(x)
     if isinstance(x, list):
         try:
-            return ops.convert_to_tensor(x)
-        except:
-            return tf.ragged.constant(x)
+            numpy_x = np.array(x)
+        except ValueError as e:
+            # If numpy conversion failed, try converting to a ragged array.
+            try:
+                return tf.ragged.constant(x)
+            except ValueError:
+                # If ragged conversion failed return to the numpy error.
+                raise e
+        # If we have a string input, use tf.tensor.
+        if numpy_x.dtype.type is np.str_:
+            return tf.convert_to_tensor(x)
+        # We have non-ragged, non-string input. Use backbend type.
+        return ops.convert_to_tensor(numpy_x)
     if is_tensor_type(x):
+        # String or ragged types we keep as tf.
+        if isinstance(x, tf.RaggedTensor) or x.dtype == tf.string:
+            return x
+        # If we have a string input, use tf.tensor.
+        if isinstance(x, np.ndarray) and x.dtype.type is np.str_:
+            return tf.convert_to_tensor(x)
         # Torch will complain about device placement for GPU tensors.
         if keras.config.backend() == "torch":
             import torch
 
             if isinstance(x, torch.Tensor):
                 x = x.cpu()
-        return x
+        return ops.convert_to_tensor(x)
     return x
 
 
-def convert_outputs(x):
+def convert_preprocessing_outputs(x):
     """Convert outputs after preprocessing to a backend agnostic format.
 
     This function is used to convert `tf.Tensor` and `tf.RaggedTensor` output
@@ -147,11 +170,11 @@ def convert_outputs(x):
     ```python
     # Two ragged arrays of token ids.
     x = tf.ragged.constant([[1, 2, 3], [4, 5]])
-    keras_nlp.utils.convert_outputs(x)
+    keras_nlp.utils.convert_preprocessing_outputs(x)
 
     # A batch of three samples each with two string segments.
     x = (tf.constant["hi", "yo", "hey"]), tf.constant(["bye", "ciao", ""]))
-    keras_nlp.utils.convert_outputs(x)
+    keras_nlp.utils.convert_preprocessing_outputs(x)
 
     # A batch of features in a dictionary.
     x = {
@@ -159,7 +182,7 @@ def convert_outputs(x):
         "images": tf.ones((3, 64, 64, 3)),
         "labels": tf.constant([1, 0, 1]),
     }
-    keras_nlp.utils.convert_outputs(x)
+    keras_nlp.utils.convert_preprocessing_outputs(x)
     ```
     """
     if not tf.executing_eagerly() or in_no_convert_scope():
diff --git a/keras_nlp/src/utils/tensor_utils_test.py b/keras_nlp/src/utils/tensor_utils_test.py
index 1eea9ff0da..c0f34595c2 100644
--- a/keras_nlp/src/utils/tensor_utils_test.py
+++ b/keras_nlp/src/utils/tensor_utils_test.py
@@ -19,8 +19,8 @@
 
 from keras_nlp.src.tests.test_case import TestCase
 from keras_nlp.src.utils.tensor_utils import any_equal
-from keras_nlp.src.utils.tensor_utils import convert_inputs
-from keras_nlp.src.utils.tensor_utils import convert_outputs
+from keras_nlp.src.utils.tensor_utils import convert_preprocessing_inputs
+from keras_nlp.src.utils.tensor_utils import convert_preprocessing_outputs
 from keras_nlp.src.utils.tensor_utils import convert_to_ragged_batch
 from keras_nlp.src.utils.tensor_utils import is_tensor_type
 from keras_nlp.src.utils.tensor_utils import tensor_to_list
@@ -30,33 +30,33 @@ class ConvertHelpers(TestCase):
     def test_basics(self):
         inputs = ops.array([1, 2, 3])
         # Convert to tf.
-        outputs = convert_inputs(inputs)
+        outputs = convert_preprocessing_inputs(inputs)
         self.assertAllEqual(outputs, ops.array(inputs))
         # Convert from tf.
-        outputs = convert_outputs(outputs)
+        outputs = convert_preprocessing_outputs(outputs)
         self.assertTrue(is_tensor_type(outputs))
         self.assertAllEqual(outputs, inputs)
 
     def test_strings(self):
         inputs = ["one", "two"]
         # Convert to tf.
-        outputs = convert_inputs(inputs)
+        outputs = convert_preprocessing_inputs(inputs)
         self.assertIsInstance(outputs, tf.Tensor)
         self.assertAllEqual(outputs, tf.constant(inputs))
         # Convert from tf.
-        outputs = convert_outputs(outputs)
+        outputs = convert_preprocessing_outputs(outputs)
         self.assertIsInstance(outputs, list)
         self.assertEqual(outputs, inputs)
 
     def test_ragged(self):
         inputs = [np.ones((1, 3)), np.ones((1, 2))]
         # Convert to tf.
-        outputs = convert_inputs(inputs)
+        outputs = convert_preprocessing_inputs(inputs)
         self.assertIsInstance(outputs, tf.RaggedTensor)
         print(outputs, inputs)
         self.assertAllEqual(outputs, tf.ragged.constant(inputs))
         # Convert from tf.
-        outputs = convert_outputs(outputs)
+        outputs = convert_preprocessing_outputs(outputs)
         self.assertIsInstance(outputs, list)
         self.assertEqual(outputs, [[[1, 1, 1]], [[1, 1]]])
 
@@ -71,14 +71,14 @@ def test_composite(self):
             [3, 4],
         )
 
-        outputs = convert_inputs(inputs)
+        outputs = convert_preprocessing_inputs(inputs)
         self.assertIsInstance(outputs[0]["text"], tf.Tensor)
         self.assertIsInstance(outputs[0]["images"], tf.RaggedTensor)
         self.assertIsInstance(outputs[0]["ragged_ints"], tf.RaggedTensor)
         self.assertTrue(is_tensor_type(outputs[1]))
         self.assertTrue(is_tensor_type(outputs[2]))
 
-        outputs = convert_outputs(outputs)
+        outputs = convert_preprocessing_outputs(outputs)
         self.assertIsInstance(outputs[0]["text"], list)
         self.assertIsInstance(outputs[0]["images"], list)
         self.assertIsInstance(outputs[0]["ragged_ints"], list)

From ce08c39e9f285483dc36876a3414328581553764 Mon Sep 17 00:00:00 2001
From: Matt Watson <mattdangerw@gmail.com>
Date: Thu, 5 Sep 2024 15:46:05 -0700
Subject: [PATCH 4/8] Torch property fixes

---
 .../preprocessing/resizing_image_converter.py  | 18 ------------------
 .../resizing_image_converter_test.py           |  3 +--
 .../pali_gemma_causal_lm_preprocessor.py       |  2 +-
 keras_nlp/src/models/preprocessor.py           |  2 +-
 keras_nlp/src/utils/tensor_utils.py            |  3 +++
 5 files changed, 6 insertions(+), 22 deletions(-)

diff --git a/keras_nlp/src/layers/preprocessing/resizing_image_converter.py b/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
index be839b3417..3344f1a9ce 100644
--- a/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
+++ b/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
@@ -37,18 +37,9 @@ class ResizingImageConverter(ImageConverter):
             largest possible window in the image (of size `(height, width)`)
             that matches the target aspect ratio. By default
             (`crop_to_aspect_ratio=False`), aspect ratio may not be preserved.
-        pad_to_aspect_ratio: If `True`, pad the images without aspect
-            ratio distortion. When the original aspect ratio differs
-            from the target aspect ratio, the output image will be
-            evenly padded on the short side.
         interpolation: String, the interpolation method.
             Supports `"bilinear"`, `"nearest"`, `"bicubic"`,
             `"lanczos3"`, `"lanczos5"`. Defaults to `"bilinear"`.
-        fill_mode: When using `pad_to_aspect_ratio=True`, padded areas
-            are filled according to the given mode. Only `"constant"` is
-            supported at this time
-            (fill with constant value, equal to `fill_value`).
-        fill_value: Float. Padding value to use when `pad_to_aspect_ratio=True`.
         data_format: string, either `"channels_last"` or `"channels_first"`.
             The ordering of the dimensions in the inputs. `"channels_last"`
             corresponds to inputs with shape `(batch, height, width, channels)`
@@ -74,10 +65,7 @@ def __init__(
         height,
         width,
         crop_to_aspect_ratio=True,
-        pad_to_aspect_ratio=False,
         interpolation="bilinear",
-        fill_mode="constant",
-        fill_value=0.0,
         data_format=None,
         **kwargs,
     ):
@@ -88,10 +76,7 @@ def __init__(
             height,
             width,
             crop_to_aspect_ratio=crop_to_aspect_ratio,
-            pad_to_aspect_ratio=pad_to_aspect_ratio,
             interpolation=interpolation,
-            fill_mode=fill_mode,
-            fill_value=fill_value,
             data_format=data_format,
         )
 
@@ -107,9 +92,6 @@ def get_config(self):
                 "width": self.resizing.width,
                 "interpolation": self.resizing.interpolation,
                 "crop_to_aspect_ratio": self.resizing.crop_to_aspect_ratio,
-                "pad_to_aspect_ratio": self.resizing.pad_to_aspect_ratio,
-                "fill_mode": self.resizing.fill_mode,
-                "fill_value": self.resizing.fill_value,
                 "data_format": self.resizing.data_format,
             }
         )
diff --git a/keras_nlp/src/layers/preprocessing/resizing_image_converter_test.py b/keras_nlp/src/layers/preprocessing/resizing_image_converter_test.py
index 9a4b8c0c02..f96a3a3488 100644
--- a/keras_nlp/src/layers/preprocessing/resizing_image_converter_test.py
+++ b/keras_nlp/src/layers/preprocessing/resizing_image_converter_test.py
@@ -38,9 +38,8 @@ def test_config(self):
         converter = ResizingImageConverter(
             width=12,
             height=20,
-            pad_to_aspect_ratio=True,
             crop_to_aspect_ratio=False,
-            fill_value=7.0,
+            interpolation="nearest",
         )
         clone = ResizingImageConverter.from_config(converter.get_config())
         test_batch = np.random.rand(4, 10, 20, 3) * 255
diff --git a/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor.py b/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor.py
index 3521f16073..01493ef454 100644
--- a/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_causal_lm_preprocessor.py
@@ -39,7 +39,7 @@ class PaliGemmaCausalLMPreprocessor(CausalLMPreprocessor):
     def __init__(
         self,
         tokenizer,
-        image_converter,
+        image_converter=None,
         sequence_length=1024,
         add_start_token=True,
         add_end_token=True,
diff --git a/keras_nlp/src/models/preprocessor.py b/keras_nlp/src/models/preprocessor.py
index f1fb32aa72..90135a88df 100644
--- a/keras_nlp/src/models/preprocessor.py
+++ b/keras_nlp/src/models/preprocessor.py
@@ -55,7 +55,7 @@ def __init__(self, *args, **kwargs):
 
     def __setattr__(self, name, value):
         # Work around torch setattr for properties.
-        if name in ["tokenizer"]:
+        if name in ["tokenizer", "audio_converter", "image_converter"]:
             return object.__setattr__(self, name, value)
         return super().__setattr__(name, value)
 
diff --git a/keras_nlp/src/utils/tensor_utils.py b/keras_nlp/src/utils/tensor_utils.py
index 1fd2a85e37..27929fe250 100644
--- a/keras_nlp/src/utils/tensor_utils.py
+++ b/keras_nlp/src/utils/tensor_utils.py
@@ -133,6 +133,9 @@ def convert_preprocessing_inputs(x):
         # If we have a string input, use tf.tensor.
         if numpy_x.dtype.type is np.str_:
             return tf.convert_to_tensor(x)
+        # Numpy will default to int64, int32 works with more ops.
+        if numpy_x.dtype == np.int64:
+            numpy_x = numpy_x.astype(np.int32)
         # We have non-ragged, non-string input. Use backbend type.
         return ops.convert_to_tensor(numpy_x)
     if is_tensor_type(x):

From dfb5dc2e2320574f504a70c0f0102861e2ab943b Mon Sep 17 00:00:00 2001
From: Matt Watson <mattdangerw@gmail.com>
Date: Thu, 5 Sep 2024 20:07:55 -0700
Subject: [PATCH 5/8] Another fix

---
 .../models/pali_gemma/pali_gemma_backbone_test.py  |  2 +-
 keras_nlp/src/utils/tensor_utils.py                | 14 ++++++++------
 keras_nlp/src/utils/transformers/preset_loader.py  |  4 ++++
 3 files changed, 13 insertions(+), 7 deletions(-)

diff --git a/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py b/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py
index 25c4b27a6f..44d80b307d 100644
--- a/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_backbone_test.py
@@ -85,7 +85,7 @@ def test_saved_model(self):
             input_data=self.input_data,
         )
 
-    @pytest.mark.large
+    @pytest.mark.extra_large
     def test_smallest_preset(self):
         self.run_preset_test(
             cls=PaliGemmaBackbone,
diff --git a/keras_nlp/src/utils/tensor_utils.py b/keras_nlp/src/utils/tensor_utils.py
index 27929fe250..94594202ab 100644
--- a/keras_nlp/src/utils/tensor_utils.py
+++ b/keras_nlp/src/utils/tensor_utils.py
@@ -137,7 +137,11 @@ def convert_preprocessing_inputs(x):
         if numpy_x.dtype == np.int64:
             numpy_x = numpy_x.astype(np.int32)
         # We have non-ragged, non-string input. Use backbend type.
-        return ops.convert_to_tensor(numpy_x)
+        x = ops.convert_to_tensor(numpy_x)
+        # Torch will complain about device placement for GPU tensors.
+        if keras.config.backend() == "torch":
+            x = x.cpu()
+        return x
     if is_tensor_type(x):
         # String or ragged types we keep as tf.
         if isinstance(x, tf.RaggedTensor) or x.dtype == tf.string:
@@ -145,13 +149,11 @@ def convert_preprocessing_inputs(x):
         # If we have a string input, use tf.tensor.
         if isinstance(x, np.ndarray) and x.dtype.type is np.str_:
             return tf.convert_to_tensor(x)
+        x = ops.convert_to_tensor(x)
         # Torch will complain about device placement for GPU tensors.
         if keras.config.backend() == "torch":
-            import torch
-
-            if isinstance(x, torch.Tensor):
-                x = x.cpu()
-        return ops.convert_to_tensor(x)
+            x = x.cpu()
+        return x
     return x
 
 
diff --git a/keras_nlp/src/utils/transformers/preset_loader.py b/keras_nlp/src/utils/transformers/preset_loader.py
index 1a1ce928ba..7e93140979 100644
--- a/keras_nlp/src/utils/transformers/preset_loader.py
+++ b/keras_nlp/src/utils/transformers/preset_loader.py
@@ -71,3 +71,7 @@ def load_backbone(self, cls, load_weights, **kwargs):
 
     def load_tokenizer(self, cls, **kwargs):
         return self.converter.convert_tokenizer(cls, self.preset, **kwargs)
+
+    def load_image_converter(self, cls, **kwargs):
+        # TODO: set image size for pali gemma checkpoints.
+        return None

From 63f177b2fe21a4d2ca48c761e0e37e9b6f2e5f1e Mon Sep 17 00:00:00 2001
From: Matt Watson <mattdangerw@gmail.com>
Date: Mon, 9 Sep 2024 15:01:30 -0700
Subject: [PATCH 6/8] Address comments

---
 keras_nlp/src/layers/preprocessing/audio_converter.py  | 10 +++++-----
 keras_nlp/src/layers/preprocessing/image_converter.py  | 10 +++++-----
 .../layers/preprocessing/resizing_image_converter.py   |  4 ++--
 keras_nlp/src/models/preprocessor.py                   |  2 +-
 keras_nlp/src/models/task.py                           |  2 +-
 keras_nlp/src/tokenizers/tokenizer.py                  |  2 +-
 keras_nlp/src/utils/preset_utils.py                    |  4 ++--
 7 files changed, 17 insertions(+), 17 deletions(-)

diff --git a/keras_nlp/src/layers/preprocessing/audio_converter.py b/keras_nlp/src/layers/preprocessing/audio_converter.py
index efc8a83117..c552b86e04 100644
--- a/keras_nlp/src/layers/preprocessing/audio_converter.py
+++ b/keras_nlp/src/layers/preprocessing/audio_converter.py
@@ -34,7 +34,7 @@ class AudioConverter(PreprocessingLayer):
     should be instantiated via the `from_preset()` constructor, which will
     create the correct subclass of this layer for the model preset.
 
-    The layer will take a input a raw audio tensor with shape `(batch_size,
+    The layer will take as input a raw audio tensor with shape `(batch_size,
     num_samples)`, and output a preprocessed audio input for modeling. The exact
     structure of the preprocessed input will vary per model. Preprocessing
     will often include computing a spectogram of the raw audio signal.
@@ -67,10 +67,10 @@ def from_preset(
         """Instantiate a `keras_nlp.layers.AudioConverter` from a model preset.
 
         A preset is a directory of configs, weights and other file assets used
-        to save and load a pre-trained model. The `preset` can be passed as a
+        to save and load a pre-trained model. The `preset` can be passed as
         one of:
 
-        1. a built in preset identifier like `'whisper_base_en'`
+        1. a built-in preset identifier like `'whisper_base_en'`
         2. a Kaggle Models handle like
            `'kaggle://user/whisper/keras/whisper_base_en'`
         3. a Hugging Face handle like `'hf://user/whisper_base_en'`
@@ -86,7 +86,7 @@ class like `keras_nlp.models.AudioConverter.from_preset()`, or from a
         will be inferred from the config in the preset directory.
 
         Args:
-            preset: string. A built in preset identifier, a Kaggle Models
+            preset: string. A built-in preset identifier, a Kaggle Models
                 handle, a Hugging Face handle, or a path to a local directory.
             load_weights: bool. If `True`, the weights will be loaded into the
                 model architecture. If `False`, the weights will be randomly
@@ -109,7 +109,7 @@ class like `keras_nlp.models.AudioConverter.from_preset()`, or from a
         return loader.load_audio_converter(cls, **kwargs)
 
     def save_to_preset(self, preset_dir):
-        """Save tokenizer to a preset directory.
+        """Save audio converter to a preset directory.
 
         Args:
             preset_dir: The path to the local model preset directory.
diff --git a/keras_nlp/src/layers/preprocessing/image_converter.py b/keras_nlp/src/layers/preprocessing/image_converter.py
index 1807d3c492..28ace2aba3 100644
--- a/keras_nlp/src/layers/preprocessing/image_converter.py
+++ b/keras_nlp/src/layers/preprocessing/image_converter.py
@@ -34,7 +34,7 @@ class ImageConverter(PreprocessingLayer):
     should be instantiated via the `from_preset()` constructor, which will
     create the correct subclass of this layer for the model preset.
 
-    The layer will take a input a raw image tensor in the channels last or
+    The layer will take as input a raw image tensor in the channels last or
     channels first format, and output a preprocessed image input for modeling.
     The exact structure of the output will vary per model, though in most cases
     this layer will simply resize the image to the size needed by the model
@@ -70,10 +70,10 @@ def from_preset(
         """Instantiate a `keras_nlp.layers.ImageConverter` from a model preset.
 
         A preset is a directory of configs, weights and other file assets used
-        to save and load a pre-trained model. The `preset` can be passed as a
+        to save and load a pre-trained model. The `preset` can be passed as
         one of:
 
-        1. a built in preset identifier like `'pali_gemma_3b_224'`
+        1. a built-in preset identifier like `'pali_gemma_3b_224'`
         2. a Kaggle Models handle like
            `'kaggle://user/paligemma/keras/pali_gemma_3b_224'`
         3. a Hugging Face handle like `'hf://user/pali_gemma_3b_224'`
@@ -90,7 +90,7 @@ class like `keras_nlp.models.ImageConverter.from_preset()`, or from a
         inferred from the config in the preset directory.
 
         Args:
-            preset: string. A built in preset identifier, a Kaggle Models
+            preset: string. A built-in preset identifier, a Kaggle Models
                 handle, a Hugging Face handle, or a path to a local directory.
             load_weights: bool. If `True`, the weights will be loaded into the
                 model architecture. If `False`, the weights will be randomly
@@ -118,7 +118,7 @@ class like `keras_nlp.models.ImageConverter.from_preset()`, or from a
         return loader.load_image_converter(cls, **kwargs)
 
     def save_to_preset(self, preset_dir):
-        """Save tokenizer to a preset directory.
+        """Save image converter to a preset directory.
 
         Args:
             preset_dir: The path to the local model preset directory.
diff --git a/keras_nlp/src/layers/preprocessing/resizing_image_converter.py b/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
index 3344f1a9ce..b8ba682f8e 100644
--- a/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
+++ b/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
@@ -24,7 +24,7 @@ class ResizingImageConverter(ImageConverter):
 
     The `ResizingImageConverter` is a subclass of `ImageConverter` for models
     that simply need to resize image tensors before using them for modeling.
-    The layer will take a input a raw image tensor (batched or unbatched) in the
+    The layer will take as input a raw image tensor (batched or unbatched) in the
     channels last or channels first format, and output a resize tensor.
 
     Args:
@@ -40,7 +40,7 @@ class ResizingImageConverter(ImageConverter):
         interpolation: String, the interpolation method.
             Supports `"bilinear"`, `"nearest"`, `"bicubic"`,
             `"lanczos3"`, `"lanczos5"`. Defaults to `"bilinear"`.
-        data_format: string, either `"channels_last"` or `"channels_first"`.
+        data_format: String, either `"channels_last"` or `"channels_first"`.
             The ordering of the dimensions in the inputs. `"channels_last"`
             corresponds to inputs with shape `(batch, height, width, channels)`
             while `"channels_first"` corresponds to inputs with shape
diff --git a/keras_nlp/src/models/preprocessor.py b/keras_nlp/src/models/preprocessor.py
index 90135a88df..fd24ea8b47 100644
--- a/keras_nlp/src/models/preprocessor.py
+++ b/keras_nlp/src/models/preprocessor.py
@@ -138,7 +138,7 @@ def from_preset(
         """Instantiate a `keras_nlp.models.Preprocessor` from a model preset.
 
         A preset is a directory of configs, weights and other file assets used
-        to save and load a pre-trained model. The `preset` can be passed as a
+        to save and load a pre-trained model. The `preset` can be passed as
         one of:
 
         1. a built-in preset identifier like `'bert_base_en'`
diff --git a/keras_nlp/src/models/task.py b/keras_nlp/src/models/task.py
index 919d025d1a..8da47992f1 100644
--- a/keras_nlp/src/models/task.py
+++ b/keras_nlp/src/models/task.py
@@ -152,7 +152,7 @@ def from_preset(
         """Instantiate a `keras_nlp.models.Task` from a model preset.
 
         A preset is a directory of configs, weights and other file assets used
-        to save and load a pre-trained model. The `preset` can be passed as a
+        to save and load a pre-trained model. The `preset` can be passed as
         one of:
 
         1. a built-in preset identifier like `'bert_base_en'`
diff --git a/keras_nlp/src/tokenizers/tokenizer.py b/keras_nlp/src/tokenizers/tokenizer.py
index 5a0b3712e4..2f8a5c9549 100644
--- a/keras_nlp/src/tokenizers/tokenizer.py
+++ b/keras_nlp/src/tokenizers/tokenizer.py
@@ -231,7 +231,7 @@ def from_preset(
         """Instantiate a `keras_nlp.models.Tokenizer` from a model preset.
 
         A preset is a directory of configs, weights and other file assets used
-        to save and load a pre-trained model. The `preset` can be passed as a
+        to save and load a pre-trained model. The `preset` can be passed as
         one of:
 
         1. a built-in preset identifier like `'bert_base_en'`
diff --git a/keras_nlp/src/utils/preset_utils.py b/keras_nlp/src/utils/preset_utils.py
index c4160ad0ae..21feaec0ae 100644
--- a/keras_nlp/src/utils/preset_utils.py
+++ b/keras_nlp/src/utils/preset_utils.py
@@ -659,11 +659,11 @@ def load_tokenizer(self, cls, **kwargs):
         raise NotImplementedError
 
     def load_audio_converter(self, cls, **kwargs):
-        """Load a audio converter layer from the preset."""
+        """Load an audio converter layer from the preset."""
         raise NotImplementedError
 
     def load_image_converter(self, cls, **kwargs):
-        """Load a image converter layer from the preset."""
+        """Load an image converter layer from the preset."""
         raise NotImplementedError
 
     def load_task(self, cls, load_weights, load_task_extras, **kwargs):

From fa248619eef369b2f113c7f03ccf42208ffcc5b0 Mon Sep 17 00:00:00 2001
From: Matt Watson <mattdangerw@gmail.com>
Date: Mon, 9 Sep 2024 16:26:19 -0700
Subject: [PATCH 7/8] Add assets on kaggle; bump preset versions

---
 keras_nlp/api/layers/__init__.py              |  3 +++
 keras_nlp/api/models/__init__.py              |  3 ---
 .../preprocessing/audio_converter_test.py     |  4 ++--
 .../preprocessing/resizing_image_converter.py |  1 -
 .../models/pali_gemma/pali_gemma_presets.py   | 10 +++++-----
 .../models/whisper/whisper_audio_converter.py |  5 ++++-
 .../src/models/whisper/whisper_backbone.py    |  3 ---
 .../models/whisper/whisper_backbone_test.py   |  1 -
 .../src/models/whisper/whisper_presets.py     | 20 +++++++++----------
 9 files changed, 24 insertions(+), 26 deletions(-)

diff --git a/keras_nlp/api/layers/__init__.py b/keras_nlp/api/layers/__init__.py
index 6431128b4e..8b92cc11b0 100644
--- a/keras_nlp/api/layers/__init__.py
+++ b/keras_nlp/api/layers/__init__.py
@@ -53,3 +53,6 @@
 from keras_nlp.src.models.pali_gemma.pali_gemma_image_converter import (
     PaliGemmaImageConverter,
 )
+from keras_nlp.src.models.whisper.whisper_audio_converter import (
+    WhisperAudioConverter,
+)
diff --git a/keras_nlp/api/models/__init__.py b/keras_nlp/api/models/__init__.py
index 7f752d1f02..64368e4c45 100644
--- a/keras_nlp/api/models/__init__.py
+++ b/keras_nlp/api/models/__init__.py
@@ -228,9 +228,6 @@
 from keras_nlp.src.models.text_classifier_preprocessor import (
     TextClassifierPreprocessor,
 )
-from keras_nlp.src.models.whisper.whisper_audio_converter import (
-    WhisperAudioConverter,
-)
 from keras_nlp.src.models.whisper.whisper_backbone import WhisperBackbone
 from keras_nlp.src.models.whisper.whisper_tokenizer import WhisperTokenizer
 from keras_nlp.src.models.xlm_roberta.xlm_roberta_backbone import (
diff --git a/keras_nlp/src/layers/preprocessing/audio_converter_test.py b/keras_nlp/src/layers/preprocessing/audio_converter_test.py
index 7e271e757f..7ac6ab22fd 100644
--- a/keras_nlp/src/layers/preprocessing/audio_converter_test.py
+++ b/keras_nlp/src/layers/preprocessing/audio_converter_test.py
@@ -65,5 +65,5 @@ def test_save_to_preset(self):
 
         # Check loading.
         restored = AudioConverter.from_preset(save_dir)
-        test_image = np.random.rand(100, 100, 3) * 255
-        self.assertAllClose(restored(test_image), converter(test_image))
+        test_audio = np.random.rand(1_000)
+        self.assertAllClose(restored(test_audio), converter(test_audio))
diff --git a/keras_nlp/src/layers/preprocessing/resizing_image_converter.py b/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
index b8ba682f8e..8012acf5c2 100644
--- a/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
+++ b/keras_nlp/src/layers/preprocessing/resizing_image_converter.py
@@ -92,7 +92,6 @@ def get_config(self):
                 "width": self.resizing.width,
                 "interpolation": self.resizing.interpolation,
                 "crop_to_aspect_ratio": self.resizing.crop_to_aspect_ratio,
-                "data_format": self.resizing.data_format,
             }
         )
         return config
diff --git a/keras_nlp/src/models/pali_gemma/pali_gemma_presets.py b/keras_nlp/src/models/pali_gemma/pali_gemma_presets.py
index c8791f709b..8cabcc7038 100644
--- a/keras_nlp/src/models/pali_gemma/pali_gemma_presets.py
+++ b/keras_nlp/src/models/pali_gemma/pali_gemma_presets.py
@@ -25,7 +25,7 @@
             "path": "pali_gemma",
             "model_card": "https://www.kaggle.com/models/google/paligemma",
         },
-        "kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_mix_224/1",
+        "kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_mix_224/2",
     },
     "pali_gemma_3b_mix_448": {
         "metadata": {
@@ -37,7 +37,7 @@
             "path": "pali_gemma",
             "model_card": "https://www.kaggle.com/models/google/paligemma",
         },
-        "kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_mix_448/1",
+        "kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_mix_448/2",
     },
     "pali_gemma_3b_224": {
         "metadata": {
@@ -49,7 +49,7 @@
             "path": "pali_gemma",
             "model_card": "https://www.kaggle.com/models/google/paligemma",
         },
-        "kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_224/1",
+        "kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_224/2",
     },
     "pali_gemma_3b_448": {
         "metadata": {
@@ -61,7 +61,7 @@
             "path": "pali_gemma",
             "model_card": "https://www.kaggle.com/models/google/paligemma",
         },
-        "kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_448/1",
+        "kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_448/2",
     },
     "pali_gemma_3b_896": {
         "metadata": {
@@ -73,6 +73,6 @@
             "path": "pali_gemma",
             "model_card": "https://www.kaggle.com/models/google/paligemma",
         },
-        "kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_896/1",
+        "kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_896/2",
     },
 }
diff --git a/keras_nlp/src/models/whisper/whisper_audio_converter.py b/keras_nlp/src/models/whisper/whisper_audio_converter.py
index 8adae7a30a..ac15be457a 100644
--- a/keras_nlp/src/models/whisper/whisper_audio_converter.py
+++ b/keras_nlp/src/models/whisper/whisper_audio_converter.py
@@ -17,6 +17,7 @@
 
 from keras_nlp.src.api_export import keras_nlp_export
 from keras_nlp.src.layers.preprocessing.audio_converter import AudioConverter
+from keras_nlp.src.models.whisper.whisper_backbone import WhisperBackbone
 
 try:
     import tensorflow as tf
@@ -24,7 +25,7 @@
     tf = None
 
 
-@keras_nlp_export("keras_nlp.models.WhisperAudioConverter")
+@keras_nlp_export("keras_nlp.layers.WhisperAudioConverter")
 class WhisperAudioConverter(AudioConverter):
     """Whisper audio converter layer.
 
@@ -66,6 +67,8 @@ class WhisperAudioConverter(AudioConverter):
     ```
     """
 
+    backbone_cls = WhisperBackbone
+
     def __init__(
         self,
         num_mels=80,
diff --git a/keras_nlp/src/models/whisper/whisper_backbone.py b/keras_nlp/src/models/whisper/whisper_backbone.py
index 5629e05f2d..e104494b2b 100644
--- a/keras_nlp/src/models/whisper/whisper_backbone.py
+++ b/keras_nlp/src/models/whisper/whisper_backbone.py
@@ -24,7 +24,6 @@
 from keras_nlp.src.models.backbone import Backbone
 from keras_nlp.src.models.whisper.whisper_decoder import WhisperDecoder
 from keras_nlp.src.models.whisper.whisper_encoder import WhisperEncoder
-from keras_nlp.src.utils.tensor_utils import assert_tf_backend
 
 
 def whisper_kernel_initializer(stddev=0.02):
@@ -117,8 +116,6 @@ def __init__(
         dtype=None,
         **kwargs,
     ):
-        assert_tf_backend(self.__class__.__name__)
-
         # === Layers ===
         self.encoder_conv_layer_1 = keras.layers.Conv1D(
             filters=hidden_dim,
diff --git a/keras_nlp/src/models/whisper/whisper_backbone_test.py b/keras_nlp/src/models/whisper/whisper_backbone_test.py
index 0b34d95cce..33bc0e2871 100644
--- a/keras_nlp/src/models/whisper/whisper_backbone_test.py
+++ b/keras_nlp/src/models/whisper/whisper_backbone_test.py
@@ -19,7 +19,6 @@
 from keras_nlp.src.tests.test_case import TestCase
 
 
-@pytest.mark.tf_only
 class WhisperBackboneTest(TestCase):
     def setUp(self):
         self.init_kwargs = {
diff --git a/keras_nlp/src/models/whisper/whisper_presets.py b/keras_nlp/src/models/whisper/whisper_presets.py
index 6684ac793f..1a7844bbbc 100644
--- a/keras_nlp/src/models/whisper/whisper_presets.py
+++ b/keras_nlp/src/models/whisper/whisper_presets.py
@@ -25,7 +25,7 @@
             "path": "whisper",
             "model_card": "https://github.com/openai/whisper/blob/main/model-card.md",
         },
-        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_tiny_en/2",
+        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_tiny_en/3",
     },
     "whisper_base_en": {
         "metadata": {
@@ -38,7 +38,7 @@
             "path": "whisper",
             "model_card": "https://github.com/openai/whisper/blob/main/model-card.md",
         },
-        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_base_en/2",
+        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_base_en/3",
     },
     "whisper_small_en": {
         "metadata": {
@@ -51,7 +51,7 @@
             "path": "whisper",
             "model_card": "https://github.com/openai/whisper/blob/main/model-card.md",
         },
-        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_small_en/2",
+        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_small_en/3",
     },
     "whisper_medium_en": {
         "metadata": {
@@ -64,7 +64,7 @@
             "path": "whisper",
             "model_card": "https://github.com/openai/whisper/blob/main/model-card.md",
         },
-        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_medium_en/2",
+        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_medium_en/3",
     },
     "whisper_tiny_multi": {
         "metadata": {
@@ -77,7 +77,7 @@
             "path": "whisper",
             "model_card": "https://github.com/openai/whisper/blob/main/model-card.md",
         },
-        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_tiny_multi/2",
+        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_tiny_multi/3",
     },
     "whisper_base_multi": {
         "metadata": {
@@ -90,7 +90,7 @@
             "path": "whisper",
             "model_card": "https://github.com/openai/whisper/blob/main/model-card.md",
         },
-        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_base_multi/2",
+        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_base_multi/3",
     },
     "whisper_small_multi": {
         "metadata": {
@@ -103,7 +103,7 @@
             "path": "whisper",
             "model_card": "https://github.com/openai/whisper/blob/main/model-card.md",
         },
-        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_small_multi/2",
+        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_small_multi/3",
     },
     "whisper_medium_multi": {
         "metadata": {
@@ -116,7 +116,7 @@
             "path": "whisper",
             "model_card": "https://github.com/openai/whisper/blob/main/model-card.md",
         },
-        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_medium_multi/2",
+        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_medium_multi/3",
     },
     "whisper_large_multi": {
         "metadata": {
@@ -129,7 +129,7 @@
             "path": "whisper",
             "model_card": "https://github.com/openai/whisper/blob/main/model-card.md",
         },
-        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_large_multi/2",
+        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_large_multi/3",
     },
     "whisper_large_multi_v2": {
         "metadata": {
@@ -143,6 +143,6 @@
             "path": "whisper",
             "model_card": "https://github.com/openai/whisper/blob/main/model-card.md",
         },
-        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_large_multi_v2/2",
+        "kaggle_handle": "kaggle://keras/whisper/keras/whisper_large_multi_v2/3",
     },
 }

From 5a1b8ecc8953ab6d3772a08e419f6b2d39788483 Mon Sep 17 00:00:00 2001
From: Matt Watson <mattdangerw@gmail.com>
Date: Mon, 9 Sep 2024 17:19:20 -0700
Subject: [PATCH 8/8] Fix last failing test

---
 .../preprocessing/image_converter_test.py     | 22 ++++++++++++++-----
 1 file changed, 17 insertions(+), 5 deletions(-)

diff --git a/keras_nlp/src/layers/preprocessing/image_converter_test.py b/keras_nlp/src/layers/preprocessing/image_converter_test.py
index d5b25f5840..c3ecefead0 100644
--- a/keras_nlp/src/layers/preprocessing/image_converter_test.py
+++ b/keras_nlp/src/layers/preprocessing/image_converter_test.py
@@ -19,7 +19,9 @@
 import pytest
 
 from keras_nlp.src.layers.preprocessing.image_converter import ImageConverter
-from keras_nlp.src.models.backbone import Backbone
+from keras_nlp.src.models.pali_gemma.pali_gemma_backbone import (
+    PaliGemmaBackbone,
+)
 from keras_nlp.src.models.pali_gemma.pali_gemma_image_converter import (
     PaliGemmaImageConverter,
 )
@@ -55,10 +57,20 @@ def test_save_to_preset(self):
             interpolation="nearest",
         )
         converter.save_to_preset(save_dir)
-        # Save a backbone so the preset is valid.
-        backbone = Backbone.from_preset(
-            "pali_gemma_3b_mix_224",
-            load_weights=False,
+        # Save a tiny backbone so the preset is valid.
+        backbone = PaliGemmaBackbone(
+            vocabulary_size=100,
+            image_size=224,
+            num_layers=1,
+            num_query_heads=1,
+            num_key_value_heads=1,
+            hidden_dim=8,
+            intermediate_dim=16,
+            head_dim=8,
+            vit_patch_size=14,
+            vit_num_heads=1,
+            vit_hidden_dim=8,
+            vit_num_layers=1,
         )
         backbone.save_to_preset(save_dir)