Skip to content

Enhance QDQ patching for marlin_gemm and moe_wna16_marlin_gemm with i… - #2615

Merged
xin3he merged 1 commit into
mainfrom
xinhe/support_vllm_0.31
Oct 16, 2026
Merged

xin3he merged 1 commit into
mainfrom
xinhe/support_vllm_0.31

Conversation

@xin3he

@xin3he xin3he commented Oct 15, 2026 •

Copy link
Copy Markdown
Contributor

…mproved parameter handling and custom op support

Type of Change

feature/bug fix/documentation/validation/others

Description

For vllm==0.31.0, _patch_moe_marlin_gemm.._patched() missing 3 required positional arguments: 'num_tokens_past_padded', 'topk_weights', 'is_k_full'。

VLLM_QDQ 插件里 patch 的函数签名和当前 vLLM 的 moe_marlin_gemm 对不上,需要在 vllm-qdq-plugin 里把签名改成和当前 vLLM 一致。

Expected Behavior & Potential Risk

the expected behavior that triggered by this PR

How has this PR been tested?

how to reproduce the test (including hardware information)

Dependency Change?

any library dependency introduced or removed

…mproved parameter handling and custom op support

Signed-off-by: Xin He <xin3.he@intel.com>
@xin3he
xin3he requested review from yiliu30 and a balanced review from Copilot October 15, 2026 02:12

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

🔵 Needs a closer look

Signature compatibility and exactly-once QDQ through custom-op interception remain unvalidated by regression tests.

2 open findings
What changed in this PR

Updates the vLLM QDQ plugin to accommodate vLLM 0.31 Marlin signatures and dense custom-op execution.

Changes:

  • Replaces fixed wrappers with signature-aware argument handling and legacy fallbacks.
  • Adds dense custom-op interception, selecting one patch layer to avoid duplicate QDQ.
File Description
benchmark/​vllm-qdq-plugin/​src/​vllm_qdq_plugin/​patch.py Adds adaptive wrappers and dense custom-op patch selection.

🧠 Review effort: Balanced


💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Comment thread benchmark/vllm-qdq-plugin/src/vllm_qdq_plugin/patch.py
Comment thread benchmark/vllm-qdq-plugin/src/vllm_qdq_plugin/patch.py
@xin3he
xin3he merged commit 123f1d4 into main Oct 16, 2026
15 checks passed
@xin3he
xin3he deleted the xinhe/support_vllm_0.31 branch October 16, 2026 08:29
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