add quantization_w4a4_fp4 qwen3 example

[wangwenmingaa]

SUMMARY:
add quantization_w4a4_fp4 qwen3 example

TEST PLAN:
"By running the qwen3_example.py file, you can compress the Qwen3-32B
model using the NVFP4 compression 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 @wangwenmingaa, 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, runnable example script that demonstrates how to apply 4-bit weight and 4-bit activation quantization (W4A4) using the NVFP4 scheme to a large language model, specifically the Qwen3-32B. This addition aims to provide users with a clear, practical guide for compressing models with this specific quantization method within the llmcompressor framework.

Highlights

• New Example Script: I've added a new example script, qwen3_example.py, under examples/quantization_w4a4_fp4/. This script demonstrates how to apply quantization to a large language model.
• Quantization Method: The example specifically showcases w4a4_fp4 quantization using the NVFP4 scheme. This involves quantizing weights to FP4 with per-group 16 and calibrating a global scale for on-the-fly FP4 activation quantization.
• Model and Dataset Usage: The script utilizes the Qwen/Qwen3-32B model and the HuggingFaceH4/ultrachat_200k dataset for calibration, providing a concrete and runnable use case for users interested in compressing this specific model.

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 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 or fill out our survey 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

This pull request adds a new example script for quantizing the Qwen3-32B model using the NVFP4 scheme. The script is well-structured. I've identified one issue regarding the output directory for saving the quantized model. Once this is addressed, the example should function as expected.

[gemini-code-assist]

The output_dir is set to MODEL_ID, but SAVE_DIR (defined on line 60) seems intended for this purpose. Using SAVE_DIR for output_dir will save the model to a directory like Qwen3-32B-NVFP4, aligning the code with the comment on line 59.

Suggested change

	output_dir=MODEL_ID,
	output_dir=SAVE_DIR,

[dsikka]

Thank you for your PR! Were you able to us this sript to produce a model? Do you mind sharing the model through the huggingface-hub?

[dsikka]

we likely do not need this many calibration samples. i would suggest <100

[dsikka]

suggested pathway is to not use output dir and save the model and tokenizer explicitly using save_pretrained