[bugfix][npugraph_ex]duplicate pattern issue

[ChenCangtao]

What this PR does / why we need it?

When the draft model also uses vllmbackend for graph compilation, the fusion pass registration occurs again, resulting in errors due to duplicate patterns.

Does this PR introduce any user-facing change?

How was this patch tested?

• vLLM version: v0.15.0
• vLLM main: https://github.com/vllm-project/vllm/commit/v0.15.0

[gemini-code-assist]

Summary of Changes

Hello @ChenCangtao, 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 resolves an issue where graph fusion patterns were being redundantly registered, leading to errors, especially when multiple compilation passes or draft models were active. By implementing a centralized and guarded registration mechanism, the system now ensures that each fusion pattern is registered only once, improving stability and preventing runtime exceptions related to duplicate pattern definitions in the graph compilation process.

Highlights

• Duplicate Pattern Prevention: Introduced a new utility function check_and_register_fusion_pass to prevent RuntimeError caused by attempting to register the same graph fusion pattern multiple times.
• Refactored Fusion Pass Registration: Modified the initialization of GraphEXMatmulAllReduceAddRMSNormPass, GraphEXAddRMSNormQuantPass, and GraphEXQKNormRopeFusionPass to utilize the new check_and_register_fusion_pass for safe pattern registration.

🧠 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

• vllm_ascend/compilation/npugraph_ex_passes/graphex_allreduce_rmsnorm_fusion_pass.py
  • Imported check_and_register_fusion_pass from npugraph_ex_utils_check.
  • Updated the __init__ method of GraphEXMatmulAllReduceAddRMSNormPass to use check_and_register_fusion_pass for registering GraphEXMiddleLayerMatmulAllReduceAddRMSNormPattern and GraphEXLastLayerMatmulAllReduceAddRMSNormPattern.
• vllm_ascend/compilation/npugraph_ex_passes/graphex_norm_quant_fusion_pass.py
  • Imported check_and_register_fusion_pass from npugraph_ex_utils_check.
  • Updated the __init__ method of GraphEXAddRMSNormQuantPass to use check_and_register_fusion_pass for registering various GraphEXAddRMSNormQuantPattern types.
• vllm_ascend/compilation/npugraph_ex_passes/graphex_qknorm_rope_fusion_pass.py
  • Imported check_and_register_fusion_pass from npugraph_ex_utils_check.
  • Updated the __init__ method of GraphEXQKNormRopeFusionPass to use check_and_register_fusion_pass for registering GraphEXQKNormRopeFusionPattern and GraphEXQKNormRopeFusionPatternWithBias.
• vllm_ascend/compilation/npugraph_ex_passes/utils/npugraph_ex_utils_check.py
  • Imported VllmConfig.
  • Added a global _register_patterns set to track registered patterns.
  • Implemented check_and_register_fusion_pass function, which attempts to register a pattern only if it hasn't been registered before, catching and logging RuntimeError for duplicate patterns.

Activity

• The pull request was opened by ChenCangtao to address an issue where duplicate pattern registrations caused errors during graph compilation, especially when a draft model used vllmbackend.

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 introduces a mechanism to prevent duplicate registration of fusion passes, which is a good approach to solve the described problem. The changes correctly apply a new helper function across several files. However, the implementation of the core helper function check_and_register_fusion_pass contains a critical typo that makes the fix ineffective. I've provided a comment with a suggested fix for this issue.

[gemini-code-assist]

There's a typo in the global variable name _resgister_patterns on lines 61 and 64. It should be _register_patterns. This typo causes the check for already registered patterns to always fail because it refers to a different, empty set. This defeats the purpose of this function and will not prevent duplicate registration errors. This is a critical bug.

Suggested change

	def check_and_register_fusion_pass(pattern_class: type, **kwargs):
	global _resgister_patterns
	eps = kwargs.get("eps", 1e-6)
	pattern_key = str(pattern_class.__name__) + str(eps)
	if pattern_key in _resgister_patterns:
	return
	pattern = pattern_class(**kwargs)
	try:
	pattern.register()
	_register_patterns.add(pattern_key)
	except RuntimeError as e:
	if "Duplicate pattern" in str(e):
	logger.warning(f"Pattern {pattern_class.__name__} eps {eps} has been registered")
	_register_patterns.add(pattern_key)
	else:
	raise e
	def check_and_register_fusion_pass(pattern_class: type, **kwargs):
	global _register_patterns
	eps = kwargs.get("eps", 1e-6)
	pattern_key = str(pattern_class.__name__) + str(eps)
	if pattern_key in _register_patterns:
	return
	pattern = pattern_class(**kwargs)
	try:
	pattern.register()
	_register_patterns.add(pattern_key)
	except RuntimeError as e:
	if "Duplicate pattern" in str(e):
	logger.warning(f"Pattern {pattern_class.__name__} eps {eps} has been registered")
	_register_patterns.add(pattern_key)
	else:
	raise e

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.