Adding new MoE e2e tests

[HDCharles]

SUMMARY:
adding e2e tests for MoE

also small fix so scheme: None doesn't error

we ignore all but first N layers of int4 because in e2e run on one H100 it takes about 2.5 minutes per expert layer so running all 48 layers would result in ~ 2 hours of additional test time.

TEST PLAN:
in progress: https://github.com/neuralmagic/llm-compressor-testing/actions/runs/19169597134

local (after disabling the cadence skip)
export TEST_DATA_FILE="${REPOS}/llm-compressor/tests/e2e/vLLM/configs/fp8_dynamic_per_tensor_moe.yaml"
pytest tests/e2e/vLLM/test_vllm.py -vs 2>&1 | tee log-fp8.log

export TEST_DATA_FILE="${REPOS}/llm-compressor/tests/e2e/vLLM/configs/w4a16_grouped_quant.yaml"
pytest tests/e2e/vLLM/test_vllm.py -vs 2>&1 | tee log-int4.log

export TEST_DATA_FILE="${REPOS}/llm-compressor/tests/e2e/vLLM/configs/fp4_nvfp4_moe.yaml"
pytest tests/e2e/vLLM/test_vllm.py -vs 2>&1 | tee log-fp4.log

[gemini-code-assist]

Summary of Changes

Hello @HDCharles, 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 primarily focuses on expanding the test coverage for advanced model architectures by introducing new end-to-end tests for Mixture-of-Experts (MoE) models within the vLLM framework. These tests cover various quantization configurations, ensuring broader validation of model performance and stability. Additionally, a minor but important change was made to improve the robustness of a core utility function, allowing it to handle a wider range of model implementations without error.

Highlights

• New MoE E2E Tests: Added three new end-to-end test configurations for Mixture-of-Experts (MoE) models, specifically targeting the Qwen3-Omni-30B-A3B-Instruct model with NVFP4, FP8_DYNAMIC, and W4A16 quantization schemes.
• Robustness Improvement: Enhanced the untie_word_embeddings utility function to gracefully handle models that do not implement get_input_embeddings or get_output_embeddings, preventing potential crashes and logging a warning instead.
• Test Cadence Logic Adjustment: Temporarily commented out the pytest.skip logic in test_vllm.py that previously skipped tests based on a mismatch with the CADENCE environment variable.

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 new end-to-end (e2e) tests for MoE (Mixture of Experts) models and includes a modification to the untie_word_embeddings function to handle cases where get_input_embeddings and get_output_embeddings are not implemented. The new tests are defined in YAML configuration files, and the test_vllm.py file is updated to accommodate these new configurations. The untie_word_embeddings function is updated to handle NotImplementedError exceptions.

[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.

[kylesayrs]

In the future can you explain in the PR description things like why you're ignoring the first 20 layers? I assume this is just to cut runtime (which is a good enough reason for me)

[dsikka]

Please add generated checkpoints to your PR description

[dsikka]

Do you have a run where the e2e test passed? The job you linked failed

[HDCharles]

Do you have a run where the e2e test passed? The job you linked failed

not yet, failures are due to being unable to upload the model, trying to get it to run

[brian-dellabetta]

nice! LGTM pending other comments