Move GPAM code to use pure Keras3

[wsxrdv]

This enables other backends (JAX, Pytorch, TensorFlow). See https://keras.io/getting_started/#configuring-your-backend

Either set

KERAS_BACKEND="jax"

or change your configuration in ~/.keras/keras.json.

[gemini-code-assist]

Summary of Changes

Hello @wsxrdv, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request refactors the GPAM model implementation to exclusively use the Keras3 API, moving away from TensorFlow-specific components. This fundamental change is crucial for allowing the model to run seamlessly across various Keras3 backends, including JAX, PyTorch, and TensorFlow, thereby enhancing its flexibility and portability without altering its core functionality.

Highlights

• Keras3 Migration: The GPAM model code has been fully migrated from TensorFlow Keras to pure Keras3 API, enabling multi-backend support.
• Backend Agnostic Operations: All TensorFlow-specific operations (e.g., tf.reshape, tf.cast, tf.einsum) have been replaced with their backend-agnostic keras.ops equivalents.
• Updated Type Hinting: Tensor type hints have been updated from tensorflow.Tensor to keras.src.backend.KerasTensor for consistency with the Keras3 API.
• Initializer Modernization: TensorFlow initializers like tf.constant_initializer and tf.random_normal_initializer have been replaced with keras.initializers.Constant and keras.initializers.RandomNormal.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Changelog

• scaaml/models/gpam.py
  • Replaced TensorFlow imports with Keras and NumPy imports.
  • Updated Rescale layer's call method input and return type hints to KerasTensor.
  • Migrated tf.constant_initializer to keras.initializers.Constant.
  • Replaced TensorFlow mathematical operations (tf.reduce_mean, tf.math.square, tf.math.rsqrt) with keras.ops equivalents.
  • Updated type hints for initializers and tensor inputs in clone_initializer and rope functions.
  • Modified tensor shape retrieval from x.shape.as_list() to list(x.shape) to accommodate different backend behaviors.
  • Converted numerous TensorFlow tensor manipulation and mathematical functions within the rope and toeplitz_matrix_rope functions to their keras.ops counterparts.
  • Changed tf.keras.layers.Activation to keras.layers.Activation.
  • Replaced TensorFlow split, einsum, and unstack operations with keras.ops equivalents in the call method of the attention mechanism.
  • Updated weight and zeros initializers to use keras.initializers.RandomNormal and keras.initializers.Zeros respectively.
  • Adjusted type hints for x in _make_head and create_heads_outputs functions to KerasTensor.

Activity

• No specific activity recorded since the pull request creation.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[wsxrdv]

Do not submit until all GPAM config experiments have been verified.

[gemini-code-assist]

Code Review

This pull request effectively migrates the GPAM model from using TensorFlow-specific APIs to pure Keras 3, which is a great step towards enabling multi-backend support. The changes are thorough and correctly replace TensorFlow functions with their Keras counterparts. I've made one suggestion to improve the readability of a tensor splitting operation by using slicing instead of keras.ops.split, which has different semantics from the original tf.split. Overall, this is a solid contribution.

[gemini-code-assist]

While using keras.ops.split is functionally correct, it's less readable than the original tf.split because it uses split indices rather than chunk sizes. For better clarity and maintainability, consider using tensor slicing, which is more explicit about the intended chunk sizes and avoids potential confusion with the different split function semantics between TensorFlow and Keras/NumPy.

        u = uv[..., :self.expand_dim]
        v = uv[..., self.expand_dim:self.expand_dim + self.expand_dim]
        base = uv[..., self.expand_dim + self.expand_dim:]