[hybrid-model] clean up and consolidate redundant fields in RadixLinearAttention

[zminglei]

Motivation

clean up and consolidate redundant fields in RadixLinearAttention

Modifications

Accuracy Tests

Qwen3-Next

python3 -m sglang.launch_server --model-path /shared/public/elr-models/Qwen/Qwen3-Next-80B-A3B-Instruct/9c7f2fbe84465e40164a94cc16cd30b6999b0cc7/ --tp 4

python benchmark/gsm8k/bench_sglang.py --data-path /shared/public/data/gsm8k/test.jsonl
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 200/200 [00:18<00:00, 10.73it/s]
Accuracy: 0.960
Invalid: 0.000
Latency: 18.734 s
Output throughput: 1695.319 token/s

Kimi Linear

python3 -m sglang.launch_server --model-path /shared/public/elr-models/moonshotai/Kimi-Linear-48B-A3B-Instruct --tp 4 --trust-remote-code

python benchmark/gsm8k/bench_sglang.py --data-path /shared/public/data/gsm8k/test.jsonl
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 200/200 [00:19<00:00, 10.10it/s]
Accuracy: 0.900
Invalid: 0.000
Latency: 19.996 s
Output throughput: 1002.499 token/s

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 @zminglei, 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 focuses on enhancing the RadixLinearAttention module by eliminating redundant fields and simplifying dimension management. The changes aim to improve code clarity and maintainability within the attention mechanism's implementation, ensuring that key and value dimensions are derived more directly and consistently across the codebase.

Highlights

• Code Simplification in RadixLinearAttention: Removed redundant dimension-related fields such as attention_tp_size, key_dim, head_k_dim, and others from the RadixLinearAttention class, consolidating dimension calculations into qk_dim and value_dim for a cleaner and more direct approach.
• Backend Logic Streamlining: Updated hybrid_linear_attn_backend.py to reflect these changes, utilizing the newly consolidated qk_dim and head_qk_dim for QKV splitting and reshaping operations in both forward_decode and forward_extend methods.
• Constructor Parameter Update: The RadixLinearAttention constructor call in qwen3_next.py was updated to remove the now-obsolete attention_tp_size parameter, aligning with the changes in the RadixLinearAttention class definition.

🧠 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. ↩

[zminglei]

/tag-and-rerun-ci retry

[gemini-code-assist]

Code Review

The pull request successfully cleans up and consolidates redundant fields within the RadixLinearAttention class and its usage in hybrid_linear_attn_backend.py and qwen3_next.py. This refactoring improves code clarity and maintainability by removing unnecessary intermediate variables and directly calculating or retrieving the required dimensions. All changes are consistent with the stated motivation of the pull request.

[hebiao064]

(not related to this line)

would you pls also move self.conv_state_len = self.conv_states_shape[-1] to attn backend __init__ function to avoid this operation for every forward?

[hebiao064]

softplus_beta and softplus_threshold should be a config read from model config, and if not existed, we can just these two magic numbre: 1.0 and 20.0

[zminglei]

/rerun-failed-ci again