MLU Check Breaks Trainer at Import Time for Older Accelerate Versions

[alex-jw-brooks]

System Info

• transformers version: 4.44.0.dev0
• Platform: Linux-4.15.0-206-generic-x86_64-with-glibc2.27
• Python version: 3.9.16
• Huggingface_hub version: 0.24.5
• Safetensors version: 0.4.4
• Accelerate version: 0.29.0
• PyTorch version (GPU?): 2.4.0+cu121 (True)
• Tensorflow version (GPU?): 2.15.1 (True)
• Flax version (CPU?/GPU?/TPU?): 0.7.0 (cpu)
• Jax version: 0.4.13
• JaxLib version: 0.4.13
• GPU type: Tesla V100-PCIE-32GB

Who can help?

@muellerzr @SunMarc

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

• Install the latest version of transformers with accelerate==0.21.0, the current minimum supported version.
• Try from transformers import trainer

For older versions of Accelerate, this import will throw a runtime error caused by:

ImportError: cannot import name 'is_mlu_available' from 'accelerate.utils' /Users/alexanderjbrooks/transformers/venv/lib/python3.9/site-packages/accelerate/utils/__init__.py)

Expected behavior

We should be able to import the trainer with the minimum supported version of Accelerate! I will submit a PR to this effect.

[SunMarc]

Hey @alex-jw-brooks, thanks for the report ! This should be fixed by this PR #32292 that was created a few days ago. Let us know this this fixes your issue !

[alex-jw-brooks]

Looks good to me, thanks @SunMarc! I'll close my PR and this issue