migrating multimodal_entailment.py from Keras2 to Keras3 and fixing other issues

[Humbulani1234]

Migrating the script multimodal_entailment.py from Keras2 to Keras3, and addressing various encountered errors.

The function make_bert_preprocessing_model from the script has been motivated by the Tensorflow example according to the script doc string. Both the Tensorflow example and the script's function do not work, they produce the following error:

  File "/home/humbulani/tensorflow-env/env/lib/python3.11/site-packages/tensorflow/core/function/trace_type/trace_type_builder.py", line 185, in from_value
    ndarray = value.__array__()
              ^^^^^^^^^^^^^^^^^
  File "/home/humbulani/tensorflow-env/env/lib/python3.11/site-packages/keras/src/backend/common/keras_tensor.py", line 108, in __array__
    raise ValueError(
ValueError: Exception encountered when calling layer 'tokenizer' (type KerasLayer).

A KerasTensor is symbolic: it's a placeholder for a shape an a dtype. It doesn't have any actual numerical value. You cannot convert it to a NumPy array.

The error is due to calling tokenizer with a symbolic tensor argument. Even when tried with Kerashub WordPiece tokenizer I get the same error.

Suggested fix:

• Removed the function make_bert_preprocessing_model and used a BertTextClassiferPreprocessor with weights loaded using from_preset method.

• Used KerasHub BertBackbone instead of the URL path provided to tfhub.

• Added some various changes to the script to get it to work.

• Besides the ResourceExhaustion error on my side, the script seems to be working.

[fchollet]

Thanks for the PR! Very nice work. Does it run on all backends?

[Humbulani1234]

Find the generated .ipynb and .md files.

Note:

• Due to dependence on tf.data API, the script still runs on Tensorflow. Future work will be to create a Dataloader class which is backend-agnostic.

[Humbulani1234]

Attached PR that adapts the script multimodal_entailment.py to jax and torch, i.e., make the script backend-agnostic

I have also generated the .md and .ipynb files.

[fchollet]

LGTM, thank you