[AWQ ] Qwen3-omni-30b-a3b-example

[JartX]

Example to quantize Qwen3 Omni with the AWQ method

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

Note: This is required to complete the testing suite, please only add the label once the PR is code complete and local testing has been performed.

[gemini-code-assist]

Summary of Changes

Hello @JartX, 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 introduces a new example script designed to guide users through the process of quantizing the Qwen3-Omni-30B-A3B-Instruct multimodal model using the AWQ method. The primary goal is to provide a clear, runnable demonstration of how to leverage the llmcompressor library to reduce the model's memory footprint and potentially improve inference efficiency, while maintaining performance through activation-aware weight quantization.

Highlights

• New Example Script: A new Python script, qwen3-omni-30b-a3b-instruct-example.py, has been added to demonstrate the quantization process.
• AWQ Quantization: The example showcases how to apply the Activation-aware Weight Quantization (AWQ) method to the Qwen3-Omni-30B-A3B-Instruct model.
• Model and Dataset Setup: The script loads the specified Qwen3-Omni model, utilizes the flickr30k dataset for calibration, and includes preprocessing steps for multimodal input.
• Quantization Recipe: An AWQModifier recipe is defined to configure 4-bit symmetric group-wise quantization for linear layers, with specific layers ignored to preserve model integrity.
• Model Saving: The quantized model and processor are saved to a designated output directory, enabling the use of the compressed model.

🧠 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 adds a new example script for quantizing the Qwen3-Omni-30B-A3B-Instruct model using the AWQ method. The script is well-structured and demonstrates the usage of llmcompressor for this purpose. I have a few minor suggestions to improve code cleanliness and maintainability, such as removing an unused import, a redundant configuration entry, and adding a final newline.

[brian-dellabetta]

Thanks for the contribution! One question, otherwise i think it would be good to include this in our examples. we had a number of users ask about qwen omni.

This loads up in vllm?

[JartX]

Thanks for the contribution! One question, otherwise i think it would be good to include this in our examples. we had a number of users ask about qwen omni.

This loads up in vllm?

Yes using the last vllm main, i go to upload the quant model

[brian-dellabetta]

Thanks for the contribution! One question, otherwise i think it would be good to include this in our examples. we had a number of users ask about qwen omni.
This loads up in vllm?

Yes using the last vllm main, i go to upload the quant model

@JartX awesome! please let link the checkpoint when you upload, would love to check it out

[JartX]

@brian-dellabetta https://huggingface.co/jart25/Qwen3-Omni-30B-A3B-Instruct-AWQ-W4A16

[kylesayrs]

Hi @JartX

There is already a qwen3 omni example #2125
Given that using the model with AWQ is not substantially different than using it with GPTQ (just change the modifier), I don't we need a separate example for this case.

[JartX]

Hi @kylesayrs , I can't run the model generated with GPTQ in vllm, so I quantized it in AWQ.

#2125 (comment)

Please, since this is just an example and the quantized model has also been published, and it took me a lot of work, study, and testing, I ask that you reconsider reintroducing it.

For example, GPTQ runs ConchLinearKernel for me, which also gives me an error, while with group_size 32 I can run the quantized model using llm-compressor.

Thank you very much.