Skip to content

[BugFix] Fix qk_norm optimization#6080

Merged
zhoutianzi666 merged 1 commit intoPaddlePaddle:developfrom
Sunny-bot1:fix_opt_qknorm
Jan 19, 2026
Merged

[BugFix] Fix qk_norm optimization#6080
zhoutianzi666 merged 1 commit intoPaddlePaddle:developfrom
Sunny-bot1:fix_opt_qknorm

Conversation

@Sunny-bot1
Copy link
Collaborator

@Sunny-bot1 Sunny-bot1 commented Jan 18, 2026

Motivation

  • 在不使用CUDA graph的情况下调用triton kernel会带来很大的lanch开销,导致性能劣化
  • 硬件隔离

Modifications

Usage or Command

Accuracy Tests

Checklist

  • Add at least a tag in the PR title.
    • Tag list: [[FDConfig],[APIServer],[Engine], [Scheduler], [PD Disaggregation], [Executor], [Graph Optimization], [Speculative Decoding], [RL], [Models], [Quantization], [Loader], [OP], [KVCache], [DataProcessor], [BugFix], [Docs], [CI], [Optimization], [Feature], [Benchmark], [Others], [XPU], [HPU], [GCU], [DCU], [Iluvatar], [Metax]]
    • You can add new tags based on the PR content, but the semantics must be clear.
  • Format your code, run pre-commit before commit.
  • Add unit tests. Please write the reason in this PR if no unit tests.
  • Provide accuracy results.
  • If the current PR is submitting to the release branch, make sure the PR has been submitted to the develop branch, then cherry-pick it to the release branch with the [Cherry-Pick] PR tag.

@paddle-bot
Copy link

paddle-bot bot commented Jan 18, 2026

Thanks for your contribution!

@codecov-commenter
Copy link

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (develop@0e0eaa1). Learn more about missing BASE report.

Additional details and impacted files
@@            Coverage Diff             @@
##             develop    #6080   +/-   ##
==========================================
  Coverage           ?   67.45%           
==========================================
  Files              ?      350           
  Lines              ?    45320           
  Branches           ?     6967           
==========================================
  Hits               ?    30569           
  Misses             ?    12513           
  Partials           ?     2238           
Flag Coverage Δ
GPU 67.45% <100.00%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@Sunny-bot1 Sunny-bot1 changed the title [BugFix] Fix qknorm optimaization [BugFix] Fix qk_norm optimization Jan 18, 2026
@zhoutianzi666 zhoutianzi666 merged commit 9dc1c74 into PaddlePaddle:develop Jan 19, 2026
19 of 22 checks passed
LLSGYN pushed a commit to LLSGYN/FastDeploy that referenced this pull request Feb 2, 2026
K11OntheBoat added a commit that referenced this pull request Feb 5, 2026
* Support norm before rope

* [Optimization] Accelerate Qwen3 QK RMSNorm via Fused Triton Kernel (#5880)

* qk rmsnorm fused

* inplace

* glm

* fix

* add qknorm layer

* fix

* update

* fix qwen3 vl

* update rl baseline

* fix qwen3 vl moe

* test

* fix qwen vl moe rl

* fix

* fix opt qknorm (#6080)

* Support Norm before Rope

* remove extra file

---------

Co-authored-by: K11OntheBoat <“ruianmaidanglao@163.com”>
Co-authored-by: sunxin <68891411+Sunny-bot1@users.noreply.github.com>
EmmonsCurse pushed a commit that referenced this pull request Feb 5, 2026
* Support norm before rope

* [Optimization] Accelerate Qwen3 QK RMSNorm via Fused Triton Kernel (#5880)

* qk rmsnorm fused

* inplace

* glm

* fix

* add qknorm layer

* fix

* update

* fix qwen3 vl

* update rl baseline

* fix qwen3 vl moe

* test

* fix qwen vl moe rl

* fix

* fix opt qknorm (#6080)

* Support Norm before Rope

* remove extra file

---------

Co-authored-by: K11OntheBoat <“ruianmaidanglao@163.com”>
Co-authored-by: sunxin <68891411+Sunny-bot1@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants