Revert "add rmsnorm kernels support for Intel XPU"

[MekkCyber]

Reverts #41563

Reverting to check what's happening

[SunMarc]

ImportError while loading conftest '/admin/home/marc/transformers/conftest.py'.
conftest.py:27: in <module>
    from transformers.testing_utils import (
src/transformers/testing_utils.py:53: in <module>
    from transformers import Trainer
src/transformers/utils/import_utils.py:1899: in __getattr__
    module = self._get_module(self._class_to_module[name])
src/transformers/utils/import_utils.py:1929: in _get_module
    raise e
src/transformers/utils/import_utils.py:1927: in _get_module
    return importlib.import_module("." + module_name, self.__name__)
src/transformers/trainer.py:41: in <module>
    from .integrations import (
src/transformers/utils/import_utils.py:1899: in __getattr__
    module = self._get_module(self._class_to_module[name])
src/transformers/utils/import_utils.py:1929: in _get_module
    raise e
src/transformers/utils/import_utils.py:1927: in _get_module
    return importlib.import_module("." + module_name, self.__name__)
src/transformers/integrations/integration_utils.py:43: in <module>
    from .. import PreTrainedModel, TrainingArguments
src/transformers/utils/import_utils.py:1899: in __getattr__
    module = self._get_module(self._class_to_module[name])
src/transformers/utils/import_utils.py:1929: in _get_module
    raise e
src/transformers/utils/import_utils.py:1927: in _get_module
    return importlib.import_module("." + module_name, self.__name__)
src/transformers/modeling_utils.py:60: in <module>
    from .integrations.eager_paged import eager_paged_attention_forward
src/transformers/integrations/eager_paged.py:6: in <module>
    from ..generation.continuous_batching.cache import PagedAttentionCache
src/transformers/generation/continuous_batching/__init__.py:16: in <module>
    from .continuous_api import ContinuousBatchingManager, ContinuousMixin
src/transformers/generation/continuous_batching/continuous_api.py:33: in <module>
    from ...integrations.hub_kernels import load_and_register_attn_kernel
src/transformers/integrations/hub_kernels.py:141: in <module>
    register_kernel_mapping(_KERNEL_MAPPING)
../miniconda3/envs/test/lib/python3.10/site-packages/kernels/layer.py:699: in register_kernel_mapping
    feature_repos = device_repo.setdefault(device.type, device.create_repo())
../miniconda3/envs/test/lib/python3.10/site-packages/kernels/layer.py:129: in create_repo
    raise ValueError(f"Unknown device type: {self.type}")
E   ValueError: Unknown device type: xpu

not stable on kernels side, let's fix this first

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[kaixuanliu]

Hi, @SunMarc @MekkCyber The issue you mentioned above may be solved after this PR: huggingface/kernels#141, which is contributed by @YangKai0616 from our team. Can you use latest version for kernels to double check? Many thanks!

[SunMarc]

yeah we need to pin kernels version to the latest one otherwise, users who upgrade transformers might face the same issue as me. Or we need some kind of check for the different devices and ask users to upgrade. Right now it is impacting all users and not only xpu users.

[MekkCyber]

I will add a general check to upgrade before using the mapping to resolve the issue for now