fix: add kernel version

[sdiazlor]

Description

Transformers (even dev or >=5.0) is not compatible with kernels>=0.15.
It raises the following error in tests:

ImportError while loading conftest '/home/runner/work/prunatree/prunatree/pruna/tests/conftest.py'.
tests/conftest.py:6: in <module>
    from .fixtures import *  # noqa: F403, F401
    ^^^^^^^^^^^^^^^^^^^^^^^
tests/fixtures.py:9: in <module>
    from transformers import AutoModelForCausalLM, AutoModelForImageTextToText, AutoProcessor, AutoTokenizer, pipeline
../.venv/lib/python3.11/site-packages/transformers/utils/import_utils.py:2045: in __getattr__
    module = self._get_module(self._class_to_module[name])
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
../.venv/lib/python3.11/site-packages/transformers/utils/import_utils.py:2239: in _get_module
    raise e
../.venv/lib/python3.11/site-packages/transformers/utils/import_utils.py:2237: in _get_module
    return importlib.import_module("." + module_name, self.__name__)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
../.venv/lib/python3.11/site-packages/transformers/models/auto/processing_auto.py:25: in <module>
    from ...image_processing_utils import ImageProcessingMixin
../.venv/lib/python3.11/site-packages/transformers/image_processing_utils.py:23: in <module>
    from .processing_utils import ImagesKwargs, Unpack
../.venv/lib/python3.11/site-packages/transformers/processing_utils.py:79: in <module>
    from .modeling_utils import PreTrainedAudioTokenizerBase
../.venv/lib/python3.11/site-packages/transformers/modeling_utils.py:73: in <module>
    from .integrations.hub_kernels import is_kernel
../.venv/lib/python3.11/site-packages/transformers/integrations/hub_kernels.py:88: in <module>
    "cuda": LayerRepository(
../.venv/lib/python3.11/site-packages/kernels/layer/layer.py:76: in __init__
    raise ValueError("Either a revision or a version must be specified.")
E   ValueError: Either a revision or a version must be specified.

This us due to the kernel version requirement: https://github.com/huggingface/kernels/releases/tag/v0.15.1

To fix it:

• Pin the kernels version until compatibility
• Add the version for following updates

Related in prunatree: https://github.com/PrunaAI/prunatree/pull/470

No fixes btm on HF: huggingface/transformers#46291

Fixes #(issue number)

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Refactor (no functional change)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update

Testing

• I added or updated tests covering my changes
• Existing tests pass locally (uv run pytest -m "cpu and not slow")

For full setup and testing instructions, see the Contributing Guide.

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my code, especially for agent-assisted changes
• I updated the documentation where necessary

---

Thanks for contributing to Pruna! We're excited to review your work.

New to contributing? Check out our Contributing Guide for everything you need to get started.

Note:

• Draft PRs or PRs without a clear and detailed overview may be delayed.
• Please mark your PR as Ready for Review and ensure the sections above are filled out.
• Contributions that are entirely AI-generated without meaningful human review are discouraged.

[gsprochette]

Hello ! I just have some questions on the two changes. It looks super good but I need to understand what's going on before I can approve :)

[gsprochette]

I'm not super sure why we need to pin a max version if the problem was that we need to specify a version as of 15.1 and this PR is doing that. Is there still a bug even after specifying a version?
Also, why 14.1?

[gsprochette]

Ah ok I only now understood that transformers itself doesn't yet specify versions yet so we do this until they specify versions in a new release, perfect 👍

[gsprochette]

Let's just leave a comment explaining the reason of the pin I would say

[gsprochette]

I understood the answer to my question only after my first review. I left two comments:

• first is just double check on the FA3 kernel version and tests
• second is me getting confused and finally just asking for a comment :)

Is there a plan to make transformers compatible with kernels>=15.1 on the HF side? Like is there an open PR or an issue we can reference and check in the future if it's time to remove the kernels pin?

Thanks a lot for finding this bug and fixing it :)

[gsprochette]

Let's just leave a comment explaining the reason of the pin I would say