How to run accelerate with PYTORCH_ENABLE_MPS_FALLBACK

[mirodil-ml]

System Info

MacOS 

transformers>=4.35.1
datasets[audio]>=2.14.7
accelerate>=0.24.1
matplotlib
wandb
tensorboard
Cython

- `Accelerate` version: 1.2.1
- Platform: macOS-14.7.1-arm64-arm-64bit
- `accelerate` bash location: .venv/bin/accelerate
- Python version: 3.12.3
- Numpy version: 2.0.2
- PyTorch version (GPU?): 2.5.1 (False)
- PyTorch XPU available: False
- PyTorch NPU available: False
- PyTorch MLU available: False
- PyTorch MUSA available: False
- System RAM: 64.00 GB
- `Accelerate` default config:
	Not found

Information

• The official example scripts
• My own modified scripts

Tasks

• One of the scripts in the examples/ folder of Accelerate or an officially supported no_trainer script in the examples folder of the transformers repo (such as run_no_trainer_glue.py)
• My own task or dataset (give details below)

Reproduction

How to set PYTORCH_ENABLE_MPS_FALLBACK environment variable when running a script with accelerate. The accelerate is not picking up the PYTORCH_ENABLE_MPS_FALLBACK environment variable when running a script, no matter where this variable is set. I tried to set this variable in the script, in the command line and in the ./zshenv, and still PyTorch is complaining it does not see this variable.

Expected behavior

expected the PYTORCH_ENABLE_MPS_FALLBACK variable be visible in the sub-process/thread.