make cuda-only tests device-agnostic

[faaany]

What does this PR do?

As the title suggests, below are the results on XPU:

================================================ short test summary info ================================================
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_00_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_01_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_02_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_03_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_04_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_05_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_06_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_07_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_08_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_09_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_10_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_11_trl_internal_testing_tiny_random_LlamaForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_12_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_13_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_14_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_15_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_16_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_17_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_18_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_19_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_20_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_21_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_22_HuggingFaceM4_tiny_random_MistralForCausalLM
PASSED tests/slow/test_sft_slow.py::SFTTrainerSlowTester::test_sft_trainer_transformers_mp_gc_device_map_23_HuggingFaceM4_tiny_random_MistralForCausalLM
============================== 24 passed, 378 deselected, 81 warnings in 196.87s (0:03:16) ==============================

[faaany]

@qgallouedec

[lewtun]

Thanks for the clean implementation @faaany ! Since we don't currently have dedicated tests per hardware, this approach seems good to me having separate decorators like require_torch_gpu and require_torch_xpu as we do in transformers.

Can you please run make precommit to solve the code quality issues?

[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.

[faaany]

Thanks for the clean implementation @faaany ! Since we don't currently have dedicated tests per hardware, this approach seems good to me having separate decorators like require_torch_gpu and require_torch_xpu as we do in transformers.

Can you please run make precommit to solve the code quality issues?

Hi @lewtun , thanks for reviewing my PR! style fixed.