Forwarding args to the Accelerator in the Trainer class

[santiag0m]

System Info

- `transformers` version: 4.46.3
- Platform: Linux
- Accelerate version: 1.3.0

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

The issue happens when initializing a Trainer object:

from transformers.trainer import Trainer

trainer = Trainer(
    accelerator_config = {**kwargs},  # Custom Accelerator arguments
)

Expected behavior

Hey!

Is there a reason on why the accelerator_config only gets forwarded to the Accelerator args when Accelerate is not available (or less than 0.28.0)?

I would like to provide custom arguments to the accelerator when I initialize the Trainer class but it seems those arguments won't be picked up by the accelerator construction due to the conditional here:

        if is_accelerate_available("0.28.0"):
            args["dataloader_config"] = dataloader_config
        else:
            args.update(accelerator_config)


        # create accelerator object
        self.accelerator = Accelerator(**args)