Integration with transformers and vLLM

[sfc-gh-goliaro]

No description provided.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[gemini-code-assist]

Summary of Changes

Hello @sfc-gh-goliaro, 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 significantly expands the capabilities of flashinfer-bench by introducing a robust integration with the Hugging Face Transformers library. The primary goal is to allow developers and researchers to easily capture and analyze the low-level computational workloads of various LLM architectures during inference. By automatically intercepting and logging critical operations, this integration facilitates performance profiling and optimization efforts, especially for quantized models and complex architectures like Mixture of Experts. The new tooling and documentation aim to make this process accessible and verifiable.

Highlights

• New Transformers Integration: Introduced a comprehensive integration with Hugging Face Transformers, enabling automatic tracing of various Large Language Model (LLM) operations within the flashinfer-bench framework.
• Extensive Operator Tracing: Added adapters to trace key LLM operations including Attention (SDPA, Flash, Flex, Eager), RMSNorm, Rotary Position Embeddings (RoPE), Token Embeddings, Activation Functions (SiLU, GELU), Mixture of Experts (MoE), and Sampling operations (multinomial, softmax, top-k).
• Multi-Model and Multi-Dtype Support: The integration supports tracing for popular models like LLaMA-3.1 (8B, 70B), Qwen3-30B-A3B, and GPT-OSS-120B, including configurations with FP8 and MXFP4 quantization.
• Enhanced Tracing Workflow: Provided new example scripts (trace_models.py, run_all_traces.sh, verify_traces.py) and detailed documentation (README.md) to streamline the process of tracing, managing, and verifying operator workloads for Transformers models.
• Improved Attention Filter Policy: Updated the AttentionFilterPolicy to correctly deduplicate workload entries for both paged and ragged attention formats, ensuring accurate and efficient trace collection.

🧠 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.

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 comprehensive integration with the Hugging Face Transformers library, enabling the tracing of various LLM operations within flashinfer-bench. The changes include new adapter modules for attention, RMSNorm, RoPE, embedding, activation, MoE, and sampling functions, along with corresponding definition files for several LLaMA, Qwen, and GPT-OSS models. New utility scripts (run_all_traces.sh, trace_models.py, verify_traces.py) are added to facilitate tracing and verification, and the .gitignore file is updated to exclude test output. The implementation demonstrates a robust approach to patching and workload collection, including support for quantized data types like FP8 and MXFP4. Overall, this is a significant and well-executed feature addition.