fix 3 failed test cases for video_llama_3 model on Intel XPU

[kaixuanliu]

This PR fixes 3 failed test cases on Intel XPU:

1.tests/models/video_llama_3/test_modeling_video_llama_3.py::VideoLlama3IntegrationTest::test_small_model_integration_test
2.tests/models/video_llama_3/test_modeling_video_llama_3.py::VideoLlama3IntegrationTest::test_small_model_integration_test_batch_wo_ima
ge
3.tests/models/video_llama_3/test_modeling_video_llama_3.py::VideoLlama3ModelTest::test_generate_with_quant_cache 

[ydshieh]

Thank you @kaixuanliu

cc @yao-matrix for a first look 🙏

[ydshieh]

Could you explain what is the issue here that requires this new KeyError?

[yao-matrix]

same question.

[kaixuanliu]

When i run pytest -rA tests/models/video_llama_3/test_modeling_video_llama_3.py::VideoLlama3ModelTest::test_generate_with_quant_cache, it will fail and return error:

pkg_name = 'optimum.quanto', return_version = True

    def _is_package_available(pkg_name: str, return_version: bool = False) -> tuple[bool, str] | bool:
        """Check if `pkg_name` exist, and optionally try to get its version"""
        spec = importlib.util.find_spec(pkg_name)
        package_exists = spec is not None
        package_version = "N/A"
        if package_exists and return_version:
            try:
                # importlib.metadata works with the distribution package, which may be different from the import
                # name (e.g. `PIL` is the import name, but `pillow` is the distribution name)
>               distributions = PACKAGE_DISTRIBUTION_MAPPING[pkg_name]
E               KeyError: 'optimum.quanto'

src/transformers/utils/import_utils.py:56: KeyError

Here is to fix this.

[yao-matrix]

@kaixuanliu it shows 'optimum': ['optimum-quanto', 'optimum-onnx', 'optimum'], in PACKAGE_DISTRIBUTION_MAPPING, so we may need another logic to search optimum-quanto distribution in optimum package, rather than using your current logic, because the package actually there, just we find it in a wrong way in this case. @ydshieh , WDYT?

[ydshieh]

Let me check ,thank you for explaining the issue.

[kaixuanliu]

In this PR, when we catch the keyError exception, it will load optimum-quanto in package = importlib.import_module(pkg_name), where pkg_name is optimum.quanto, and it can be imported properly.

[yao-matrix]

as above

[kaixuanliu]

Done

[yao-matrix]

as above

[ydshieh]

@kaixuanliu @yao-matrix Thank you.

I have currently some trouble to get a CI runner to check the _is_package_available issue. From our scheduled CI runs, there is no issue. Of course, it's likely depending on the sys. environment, but I don't want to merge this PR without being verified more deeply myself.

I am happy to merge the PR if you can revert this part of change, and open an issue for this _is_package_available. Don't forget to provide the env. details (transformers env and pip freeze). WDYT?

[ydshieh]

run-slow: vit, clip

[github-actions]

This comment contains run-slow, running the specified jobs:

models: ["models/clip", "models/vit"]
quantizations: []

[github-actions]

CI Results

Workflow Run ⚙️

✅ No failing test specific to this PR 🎉 !

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: video_llama_3

[kaixuanliu]

@SunMarc , Hi, have reverted the code and submitted issue here: #41981

[ydshieh]

👍 🙏