Use swa radix cache and memory pool for gpt-oss model

[ispobock]

Motivation

#17220 (review)

Modifications

Accuracy Tests

|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value |   |Stderr|
|-----|------:|----------------|-----:|-----------|---|-----:|---|-----:|
|gsm8k|      3|flexible-extract|     8|exact_match|↑  |0.8893|±  |0.0086|
|     |       |strict-match    |     8|exact_match|↑  |0.3586|±  |0.0132|

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.
• Follow the SGLang code style guidance.

Review Process

1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
4. After green CI and required approvals, ask Merge Oncalls to merge.

[gemini-code-assist]

Summary of Changes

Hello @ispobock, 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 integrates GPT-OSS models with the SWA radix cache and memory pool system. The changes involve updating the model configuration logic to correctly identify and configure SWA layers for GPT-OSS based on its specific architecture. Additionally, server arguments are adjusted to manage memory behavior, including a temporary measure for the trtllm_mha attention backend. The primary goal is to enhance memory efficiency and performance for GPT-OSS models by enabling and optimizing their use of Sliding Window Attention.

Highlights

• GPT-OSS Hybrid SWA Integration: Enabled GPT-OSS models to utilize hybrid Sliding Window Attention (SWA) by including 'GptOssForCausalLM' in the list of supported architectures, allowing it to leverage SWA capabilities for improved memory efficiency.
• Dynamic SWA Layer Configuration: Updated the get_hybrid_layer_ids function to dynamically determine SWA and full attention layers for GPT-OSS models based on their layer_types configuration, providing more flexible and model-specific SWA setup.
• TRTLLM MHA Backend Compatibility: Introduced a temporary adjustment for GPT-OSS models using the trtllm_mha attention backend, setting swa_full_tokens_ratio to 1.0 and logging a warning due to current limitations in SWA KV indices translation for this specific backend.
• Hybrid SWA Memory Management: Removed the explicit disabling of hybrid SWA memory for GPT-OSS models, allowing them to fully leverage the memory pool for SWA, which is crucial for performance and memory optimization.

🧠 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 correctly enables SWA radix cache and memory pool for the GptOssForCausalLM model. The changes include updating the model configuration to recognize the new architecture and its hybrid layer structure, and adjusting server arguments to enable the feature while handling backend-specific limitations. I've identified one potential runtime error due to a missing check, which could make the server crash if the model config is missing an expected attribute. Addressing this will improve the robustness of the changes.

[ispobock]

/tag-and-rerun-ci

[ispobock]

https://github.com/sgl-project/sglang/actions/runs/21097509301/job/60697462563?pr=17261