Add guide for writing a custom kernel

[jeffcarp]

Walks users through how to define a custom kernel using Pallas (only works with the JAX backend).

[gemini-code-assist]

Summary of Changes

Hello @jeffcarp, 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 adds a comprehensive guide to the documentation, detailing the process of creating and integrating high-performance custom kernels for Keras models targeting TPUs and GPUs. The guide focuses on using the Pallas library with the JAX backend, illustrating how to achieve significant performance gains through techniques like operator fusion and memory tiling, and also covers the necessary steps to make these custom kernels trainable.

Highlights

• Custom Kernel Guide: Introduces a new guide on defining custom TPU/GPU kernels using the Pallas library for Keras models (JAX backend only).
• Pallas Integration: Demonstrates how to integrate simple Pallas kernels (e.g., vector addition) into Keras layers.
• Performance Optimization: Explores advanced concepts like operator fusion and tiling to optimize kernel performance, exemplified by a fused Matmul + ReLU layer.
• Benchmarking: Includes a benchmark comparing the performance of standard Keras operations against the custom fused Pallas kernel, showing significant speedup.
• Trainability: Explains how to enable training for custom Pallas kernels by implementing a custom backward pass using JAX's custom_vjp decorator.

🧠 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.

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. ↩

[gemini-code-assist]

Code Review

This pull request adds a comprehensive guide on creating custom TPU/GPU kernels in Keras using Pallas. The guide is well-structured, starting with a simple example and moving to a more complex, real-world use case of a fused matmul-ReLU operation, including benchmarking and enabling training. My review focuses on improving the clarity and correctness of the examples and explanations. I've suggested corrections to a misleading explanation about memory tiling, pointed out an incorrect shape in one of the training examples, and recommended cleaning up the import statements for better code organization. Overall, this is a valuable addition to the documentation.

[hertschuh]

I don't understand. The outputs are missing in your rendered colab. They should appear in blocks like this: https://github.com/keras-team/keras-io/blob/master/examples/keras_rs/md/sequential_retrieval.md?plain=1#L311

(they are correctly appearing in the ipynb file though)