[Fix][torch.compile] Enable custom ops by default when Inductor off

[ProExpertProg]

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

Purpose

This PR fixes default custom op enablement (incomplete fix in #19618). When Inductor is disabled, custom ops should be enabled by default.

Test Plan

Verified that before this PR, custom ops were disabled even when inductor was disabled. With this PR, I verified enablement works correctly.

(Optional) Documentation Update

Added some more docs in config.py

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Summary of Changes

Hello @ProExpertProg, 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 provides a fix to ensure that custom operations are correctly enabled by default when PyTorch Inductor is not being used for compilation. It refines the underlying logic that determines the default state of custom ops and updates relevant documentation to reflect this behavior, resolving an incomplete fix from a prior PR.

Highlights

• Custom Op Default Behavior: Modified the CustomOp.default_on logic in vllm/model_executor/custom_op.py to ensure custom operations are enabled by default when PyTorch Inductor is not utilized. This addresses an incomplete fix from a previous PR, making the enablement conditional on compilation_config.level < CompilationLevel.PIECEWISE OR not compilation_config.use_inductor.
• Configuration Logic Adjustment: Removed a redundant code block in vllm/config.py's CompilationConfig.__post_init__ method (lines 4472-4483). This block was explicitly setting self.compilation_config.custom_ops = ["none"] when use_inductor was true and custom_ops was empty, which inadvertently interfered with the desired default behavior of custom ops when Inductor was off.
• Documentation Clarity: Updated docstrings within vllm/config.py for the custom_ops (lines 3931-3933) and use_inductor (lines 3941-3948) fields. These updates now clearly state the default enablement/disablement of custom operations based on Inductor usage and compilation levels, improving understanding of the configuration options.

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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

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 issue 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 is currently in preview and 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 to provide feedback.

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

The pull request aims to fix the custom op enablement logic when Inductor is disabled, ensuring custom ops are enabled by default in this scenario. The changes involve modifications to vllm/config.py to update the documentation and logic related to use_inductor and custom ops, and vllm/model_executor/custom_op.py to adjust the default_on method for custom op enablement. The feedback focuses on improving clarity and consistency in the documentation and code.

[zou3519]

Code looks good to me, thank you for catching this. Could you try to add a test for this?

[zou3519]

thank you