docs: explain save vs save_weights for TFQ layers

[SripriyaDevarapalli]

Description

This pull request addresses the documentation issue in #899 regarding the behavior of tf.keras.Model.save() versus save_weights() for TensorFlow Quantum (TFQ) layers.

Summary of changes

• Documents the reported serialization issue when saving complete Keras models containing TFQ layers that may not support configuration serialization.
• Explains save_weights() as an alternative when complete model serialization fails.
• Provides a Python example showing how to save weights and load them into a recreated model architecture.
• Links to the TFQ installation guide and notes that saving and loading behavior may depend on the TensorFlow and Keras versions in use.

Related to #899 and #56.

How Has This Been Tested?

• Ran git diff --cached --check successfully.
• Reviewed the documentation changes.
• • A TFQ runtime save/load test could not be completed because the local checkout lacks the compiled native operation library required to import TFQ.
• This is a documentation-only change; no source code or tests were modified.

Test Configuration:

• Python version: 3.11
• Operating system: Windows
• TensorFlow version: 2.19.1
• tf-keras version: 2.19.0

AI use

• No AI tool was used.
• AI-assisted: I did the research and wrote the change, and used an AI tool for parts of it (for example completion, editing or review).
• AI-generated: an AI tool did most of the research or wrote most of the change. I have read all of it and can explain it.
• Translated into English with an AI tool.

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new errors or warnings
• My code and tests pass locally and have at least 80% line coverage
• Any dependent changes have been merged and published in downstream modules

[gemini-code-assist]

Code Review

This pull request updates the design documentation in docs/design.md to include a new section on saving Keras models. It explains why saving a complete model with tf.keras.Model.save() can fail for certain layers and provides a code example demonstrating how to use save_weights() and load_weights() as a workaround. There are no review comments, so there is no feedback to provide.

[SripriyaDevarapalli]

Hey @mhucka I've created a PR #1120 addressing the documentation issue regarding the behavior of save() versus save_weights() for TensorFlow Quantum layers.

The changes explain the reported serialization issue, document save_weights() as an alternative, and include a usage example, and if you have any suggestions for improvement please do let me know .

Thank you!

[SripriyaDevarapalli]

I also noticed that the CI run failed in the Tutorial tests job.

The logs show an ImportError indicating that pydot and/or Graphviz are missing when tf.keras.utils.plot_model() is called. The file format and lint checks, wheel test, and library tests all passed.

This appears to be a missing dependency in the tutorial test environment rather than an issue caused by my documentation changes. I wanted to bring this to your attention for investigation.

Thank you!