Skip to content

[RadixTree][7/N Refactor]: Refactor mamba radix tree, release dup kvcache in insert func#19429

Merged
ispobock merged 3 commits intosgl-project:mainfrom
hzh0425:hybrid/mamba-release
Mar 13, 2026
Merged

[RadixTree][7/N Refactor]: Refactor mamba radix tree, release dup kvcache in insert func#19429
ispobock merged 3 commits intosgl-project:mainfrom
hzh0425:hybrid/mamba-release

Conversation

@hzh0425
Copy link
Copy Markdown
Collaborator

@hzh0425 hzh0425 commented Feb 26, 2026

Motivation

Modifications

Accuracy Tests

python3 test/registered/4-gpu-models/test_qwen3_next_models.py

image

Benchmarking and Profiling

Checklist

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
Copy link
Copy Markdown
Contributor

Summary of Changes

Hello @hzh0425, 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 Mamba Radix Tree's memory management to enhance efficiency by intelligently handling duplicate KV cache entries. By introducing a prev_prefix_len parameter, the system can now precisely determine which parts of the KV cache to release during insertion, thereby optimizing memory usage and preventing redundant allocations. The changes are supported by new unit tests to ensure correctness and robustness.

Highlights

  • Optimized KV Cache Management: Introduced a prev_prefix_len parameter to the insert and _insert_helper functions in the Mamba Radix Tree, enabling more granular control over KV cache freeing during insertion.
  • Reduced Duplicate KV Cache Allocation: Implemented logic to conditionally free duplicate KV cache entries based on prev_prefix_len, preventing unnecessary memory allocation for shared prefixes.
  • Refactored Cache Freeing Logic: Removed redundant KV cache freeing calls from cache_finished_req and _skip_cache_unfinished_req methods, centralizing the freeing logic within _insert_helper.
  • Comprehensive Unit Testing: Added a new unit test, test_insert_prev_prefix_len, to thoroughly validate the new KV cache freeing mechanism under various scenarios, including full, partial, and no freeing.

🧠 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
  • python/sglang/srt/mem_cache/mamba_radix_cache.py
    • Modified insert method to accept prev_prefix_len in InsertParams.
    • Updated _insert_helper signature to include prev_prefix_len with a default value.
    • Implemented conditional freeing of KV cache entries within _insert_helper based on prev_prefix_len and total_prefix_length.
    • Removed unconditional token_to_kv_pool_allocator.free calls from cache_finished_req and _skip_cache_unfinished_req.
  • test/registered/radix_cache/test_mamba_unittest.py
    • Added a new helper method _setup_tree_and_allocator to streamline test environment setup for MambaRadixCache.
    • Introduced test_insert_prev_prefix_len to verify the correct behavior of KV cache freeing with different prev_prefix_len values during insertion.
Activity
  • The pull request was created by hzh0425.
  • No human activity (comments, reviews) has been recorded on this pull request yet.
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.

@hzh0425 hzh0425 changed the title [RadixTree][6/N Refactor]: Refactor mamba radix tree, release dup kvcache in insert func [RadixTree][7/N Refactor]: Refactor mamba radix tree, release dup kvcache in insert func Feb 26, 2026
Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request refactors the Mamba radix tree to correctly release duplicate KV cache entries during insertion. The core logic is moved into the _insert_helper function, which now accepts a prev_prefix_len parameter to protect a certain prefix length from being freed. This is a good refactoring that centralizes the logic and improves correctness. The accompanying new test, test_insert_prev_prefix_len, is comprehensive and validates the new behavior under various scenarios.

My main feedback is regarding code duplication in the test file. The new test helper _setup_tree_and_allocator duplicates a significant amount of setup code from an existing test. This could be improved for better maintainability.

Comment thread test/registered/radix_cache/test_mamba_unittest.py
# Conflicts:
#	python/sglang/srt/mem_cache/mamba_radix_cache.py
@ispobock ispobock merged commit 197f807 into sgl-project:main Mar 13, 2026
411 of 458 checks passed
liubiyongge pushed a commit to liubiyongge/sglang that referenced this pull request Mar 13, 2026
yhyang201 pushed a commit to yhyang201/sglang that referenced this pull request Mar 15, 2026
Wangzheee pushed a commit to Wangzheee/sglang that referenced this pull request Mar 21, 2026
0-693 pushed a commit to 0-693/sglang that referenced this pull request Mar 25, 2026
JustinTong0323 pushed a commit to JustinTong0323/sglang that referenced this pull request Apr 7, 2026
yhyang201 pushed a commit to yhyang201/sglang that referenced this pull request Apr 22, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants