System Info

- `Accelerate` version: 0.19.0
- Platform: macOS-13.3.1-arm64-arm-64bit
- Python version: 3.11.3
- Numpy version: 1.24.3
- PyTorch version (GPU?): 2.0.1 (False)
- System RAM: 32.00 GB
- `Accelerate` default config:
	- compute_environment: LOCAL_MACHINE
	- distributed_type: NO
	- mixed_precision: no
	- use_cpu: False
	- num_processes: 1
	- machine_rank: 0
	- num_machines: 1
	- rdzv_backend: static
	- same_network: True
	- main_training_function: main
	- downcast_bf16: no
	- tpu_use_cluster: False
	- tpu_use_sudo: False
	- tpu_env: []
	- dynamo_config: {'dynamo_backend': 'INDUCTOR'}

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

Logger used beforehand

Device is set to MPS, although we set cpu to True.

from accelerate import Accelerator
from accelerate.logging import get_logger

logger = get_logger(__name__)
logger.warning(f"Starting test")
accelerator = Accelerator(cpu=True)
print(f"Device: {accelerator.state.device}")

Output:

Starting test
Device: mps

Logger NOT used beforehand

Device is set to CPU, as expected.

from accelerate import Accelerator
from accelerate.logging import get_logger

logger = get_logger(__name__)
# logger.warning(f"Starting test")
accelerator = Accelerator(cpu=True)
print(f"Device: {accelerator.state.device}")

Output:

Device: cpu

Expected behavior

I would expect that whenever cpu=True then device should always be CPU. Also, if I pass a certain device it should use that device. I don't see why using the logger beforehand would have any impact on that.