What does this PR do?

Fixes a latent AttributeError in examples/t2i_adapter/train_t2i_adapter_sdxl.py that breaks --resume_from_checkpoint.

The accelerate load_model_hook references args.control_type, which is never declared as a CLI argument in this script:

def load_model_hook(models, input_dir):
    while len(models) > 0:
        model = models.pop()

        # load diffusers style into model
        load_model = T2IAdapter.from_pretrained(os.path.join(input_dir, "t2iadapter"))

        if args.control_type != "style":
            model.register_to_config(**load_model.config)

        model.load_state_dict(load_model.state_dict())
        del load_model

So any resume (accelerator.load_state(...) triggered by --resume_from_checkpoint) crashes with:

AttributeError: 'Namespace' object has no attribute 'control_type'

The args.control_type != "style" gate looks like a copy-paste leftover from a controlnet variant that distinguished style adapters. This T2IAdapter training script has no such distinction, and the analogous hook in examples/controlnet/train_controlnet_sdxl.py calls register_to_config unconditionally. Dropping the broken gate aligns with that script and makes resume work.

Fix

-                if args.control_type != "style":
-                    model.register_to_config(**load_model.config)
+                model.register_to_config(**load_model.config)

Before submitting

• Did you read the contributor guideline?

Who can review?

@sayakpaul