[opt kimi k2 2/n] apply kimi k2 thinking moe_fused_gate

[BBuf]

Motivation

Follow #13287

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.

[gemini-code-assist]

Summary of Changes

Hello @BBuf, 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 introduces a specialized and optimized CUDA kernel for handling Mixture-of-Experts (MoE) gating within the Kimi K2 model architecture. By replacing a more general implementation with a fused gate kernel tailored for Kimi K2's specific configuration, the change aims to improve the efficiency and performance of MoE operations for this model.

Highlights

• Introduction of Kimi K2 Fused Gate Kernel: A new CUDA kernel, "kimi_k2_moe_fused_gate", has been introduced and imported from "sgl_kernel" to handle Mixture-of-Experts (MoE) gating specifically for the Kimi K2 model architecture.
• Optimized Kimi K2 MoE Gating Logic: The "biased_grouped_topk_gpu" function now leverages the "kimi_k2_moe_fused_gate" for an optimized path when processing Kimi K2 models (identified by 384 experts and a single expert group), replacing the previous "kimi_k2_biased_topk_impl".
• Streamlined Kernel Interface: The new "kimi_k2_moe_fused_gate" kernel simplifies its interface by no longer requiring "hidden_states", "num_token_non_padded", and "expert_location_dispatch_info" as direct arguments, and explicitly casts "gating_output" to "float32".

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 replaces the kimi_k2_biased_topk_impl with a fused CUDA kernel kimi_k2_moe_fused_gate for performance optimization on Kimi K2 models. While the change correctly introduces the new kernel, it omits the necessary post-processing of topk_ids for logical-to-physical expert ID mapping and handling of padded tokens. This is a critical issue that could lead to incorrect expert routing. I've provided a suggestion to re-introduce this post-processing logic.