[keras-hub]Issue in running recently added models

[cosmo3769]

When I am running the recently added models in keras-hub branch, I am getting an error.

import keras_nlp
import numpy as np
input_data = np.ones((2, 224, 224, 3), dtype="float32")

model = keras_nlp.models.ViTDetBackbone.from_preset("kaggle://keras/vitdet/keras/vitdet_base/2")
model(input_data)

Traceback (most recent call last):
  File "/Users/cosmo3769/projects/keras-nlp/test.py", line 64, in <module>
    model = keras_nlp.models.ViTDetBackbone.from_preset("kaggle://keras/vitdet/keras/vitdet_base/2")
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/cosmo3769/projects/keras-nlp/keras_nlp/src/models/backbone.py", line 212, in from_preset
    if not issubclass(preset_cls, cls):
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: issubclass() arg 1 must be a class

Note: I am getting the same error in all the recently added models in keras-hub branch.

[cosmo3769]

Ohh, I think kaggle models need to be updated for this to work.

[Humbulani1234]

I tried to investigate this issue further, and here is what I found out:

The class function from_presets executes the line preset_cls = check_config_class(preset) which in turn executes the function check_config_class from keras_nlp.src.utils.preset_utils reproduced below:

def check_config_class(
    preset,
    config_file=CONFIG_FILE,
):
    """Validate a preset is being loaded on the correct class."""
    config_path = get_file(preset, config_file)
    with open(config_path, encoding="utf-8") as config_file:
        config = json.load(config_file)
    return keras.saving.get_registered_object(config["registered_name"])

The above calls the function get_registered_object from keras.src.saving.object_registration and also reproduced below:

def get_registered_object(name, custom_objects=None, module_objects=None):
    """Returns the class associated with `name` if it is registered with Keras.

    This function is part of the Keras serialization and deserialization
    framework. It maps strings to the objects associated with them for
    serialization/deserialization.

    Example:

    ```python
    def from_config(cls, config, custom_objects=None):
        if 'my_custom_object_name' in config:
            config['hidden_cls'] = tf.keras.saving.get_registered_object(
                config['my_custom_object_name'], custom_objects=custom_objects)

Args:
    name: The name to look up.
    custom_objects: A dictionary of custom objects to look the name up in.
        Generally, custom_objects is provided by the user.
    module_objects: A dictionary of custom objects to look the name up in.
        Generally, module_objects is provided by midlevel library
        implementers.

Returns:
    An instantiable class associated with `name`, or `None` if no such class
        exists.
custom_objects_scope_dict = global_state.get_global_attribute(
    "custom_objects_scope_dict", {}
)
if name in custom_objects_scope_dict:
    return custom_objects_scope_dict[name]
elif name in GLOBAL_CUSTOM_OBJECTS:
    return GLOBAL_CUSTOM_OBJECTS[name]
elif custom_objects and name in custom_objects:
    return custom_objects[name]
elif module_objects and name in module_objects:
    return module_objects[name]
return None

which in turn calls ```get_global_attribute``` from ```keras.src.backend.common.global_state``, reproduced below`:

```python
def get_global_attribute(name, default=None, set_to_default=False):
    attr = getattr(GLOBAL_STATE_TRACKER, name, None)
    if attr is None and default is not None:
        attr = default
        if set_to_default:
            set_global_attribute(name, attr)
    return attr

I suspect this is where the problem is, the line attr = getattr(GLOBAL_STATE_TRACKER, name, None) when executed returns None object which is then assigned to attr and returned to the caller get_registered_object which in turn then returns None to check_config_class. The preset_cls local variable will the be assigned the None object, hence when we execute the line if not issubclass(preset_cls, cls) from from_presets it raises the error TypeError: issubclass() arg 1 must be a class.

I also noted that the downloaded config.json file has the registered name key given by: "keras_cv.models>ViTDetBackbone". I am not quite sure about whether the > should be a ..

I hope this issue maybe be clarified further.

The following is my detailed stacktrace produced by pdb:

2024-08-27 12:41:16.184541: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-08-27 12:41:17.025270: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
> /home/humbulani/keras-nlp/keras_nlp/src/models/backbone.py(214)from_preset()
-> preset_cls = check_config_class(preset)
(Pdb) s
--Call--
> /home/humbulani/keras-nlp/keras_nlp/src/utils/preset_utils.py(582)check_config_class()
-> def check_config_class(
(Pdb) s
> /home/humbulani/keras-nlp/keras_nlp/src/utils/preset_utils.py(587)check_config_class()
-> config_path = get_file(preset, config_file)
(Pdb) n
> /home/humbulani/keras-nlp/keras_nlp/src/utils/preset_utils.py(588)check_config_class()
-> with open(config_path, encoding="utf-8") as config_file:
(Pdb) n
> /home/humbulani/keras-nlp/keras_nlp/src/utils/preset_utils.py(589)check_config_class()
-> config = json.load(config_file)
(Pdb) n
> /home/humbulani/keras-nlp/keras_nlp/src/utils/preset_utils.py(588)check_config_class()
-> with open(config_path, encoding="utf-8") as config_file:
(Pdb) n
> /home/humbulani/keras-nlp/keras_nlp/src/utils/preset_utils.py(590)check_config_class()
-> return keras.saving.get_registered_object(config["registered_name"])
(Pdb) s
--Call--
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/saving/object_registration.py(188)get_registered_object()
-> @keras_export(
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/saving/object_registration.py(222)get_registered_object()
-> custom_objects_scope_dict = global_state.get_global_attribute(
(Pdb) s
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/saving/object_registration.py(223)get_registered_object()
-> "custom_objects_scope_dict", {}
(Pdb) s
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/saving/object_registration.py(222)get_registered_object()
-> custom_objects_scope_dict = global_state.get_global_attribute(
(Pdb) s
--Call--
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/backend/common/global_state.py(15)get_global_attribute()
-> def get_global_attribute(name, default=None, set_to_default=False):
(Pdb) list
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/backend/common/global_state.py(16)get_global_attribute()
-> attr = getattr(GLOBAL_STATE_TRACKER, name, None)
(Pdb) n
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/backend/common/global_state.py(17)get_global_attribute()
-> if attr is None and default is not None:
(Pdb) p attr
None
(Pdb) n
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/backend/common/global_state.py(18)get_global_attribute()
-> attr = default
(Pdb) s
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/backend/common/global_state.py(19)get_global_attribute()
-> if set_to_default:
(Pdb) s
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/backend/common/global_state.py(21)get_global_attribute()
-> return attr
(Pdb) s
--Return--
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/backend/common/global_state.py(21)get_global_attribute()->{}
-> return attr
(Pdb) s
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/saving/object_registration.py(225)get_registered_object()
-> if name in custom_objects_scope_dict:
(Pdb) p name
'keras_cv.models>ViTDetBackbone'
(Pdb) s
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/saving/object_registration.py(227)get_registered_object()
-> elif name in GLOBAL_CUSTOM_OBJECTS:
(Pdb) s
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/saving/object_registration.py(229)get_registered_object()
-> elif custom_objects and name in custom_objects:
(Pdb) s
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/saving/object_registration.py(231)get_registered_object()
-> elif module_objects and name in module_objects:
(Pdb) s
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/saving/object_registration.py(233)get_registered_object()
-> return None
(Pdb) s
--Return--
> /home/humbulani/keras/env/lib/python3.10/site-packages/keras/src/saving/object_registration.py(233)get_registered_object()->None
-> return None
(Pdb) s
--Return--
> /home/humbulani/keras-nlp/keras_nlp/src/utils/preset_utils.py(590)check_config_class()->None
-> return keras.saving.get_registered_object(config["registered_name"])
(Pdb) s
> /home/humbulani/keras-nlp/keras_nlp/src/models/backbone.py(215)from_preset()
-> if not issubclass(preset_cls, cls):
(Pdb) p preset_cls
None
(Pdb) s
TypeError: issubclass() arg 1 must be a class
> /home/humbulani/keras-nlp/keras_nlp/src/models/backbone.py(215)from_preset()
-> if not issubclass(preset_cls, cls):
(Pdb) s
--Return--
> /home/humbulani/keras-nlp/keras_nlp/src/models/backbone.py(215)from_preset()->None
-> if not issubclass(preset_cls, cls):
(Pdb) s
TypeError: issubclass() arg 1 must be a class
> /home/humbulani/keras-nlp/test.py(5)<module>()
-> model = keras_nlp.api.models.ViTDetBackbone.from_preset("kaggle://keras/vitdet/keras/vitdet_base/2")
(Pdb) s
--Return--
> /home/humbulani/keras-nlp/test.py(5)<module>()->None
-> model = keras_nlp.api.models.ViTDetBackbone.from_preset("kaggle://keras/vitdet/keras/vitdet_base/2")
(Pdb) s
Traceback (most recent call last):
  File "/home/humbulani/keras-nlp/test.py", line 5, in <module>
--Call--

(Pdb) n
--Call--
> /usr/lib/python3.10/codecs.py(319)decode()
-> def decode(self, input, final=False):
(Pdb) n

(Pdb) n
    model = keras_nlp.api.models.ViTDetBackbone.from_preset("kaggle://keras/vitdet/keras/vitdet_base/2")
  File "/home/humbulani/keras-nlp/keras_nlp/src/models/backbone.py", line 215, in from_preset

(Pdb) c
    if not issubclass(preset_cls, cls):
TypeError: issubclass() arg 1 must be a class

[hsuyab]

I think this error is coming up in one of the example codes as well. Here in the getting started running classifier = keras_nlp.models.BertClassifier.from_preset("bert_tiny_en_uncased_sst2") also gives the TypeError: issubclass() arg 1 must be a class.

[mattdangerw]

Thanks! See the error on classifier = keras_nlp.models.BertClassifier.from_preset("bert_tiny_en_uncased_sst2") taking a look.

[mattdangerw]

Ok! This should be now working as intende on master.

classifier = keras_nlp.models.BertClassifier.from_preset("bert_tiny_en_uncased_sst2") will work as expected.

backbone = keras_nlp.models.ViTDetBackbone.from_preset("kaggle://keras/vitdet/keras/vitdet_base/2") will thrown because we are attempting to revive a KerasCV model that is not available in the package. That is expected as well. But the error message will better.

For vit det (and other models we are adding to the hub) we will need to generate new kaggle assets before the model is ready for use. That's expected.