why we must set text_encoder_three in SD3 dreambooth traning script

[ZhuangLii]

https://github.com/huggingface/diffusers/blob/5588725e8e7be497839432e5328c596169385f16/examples/dreambooth/train_dreambooth_lora_sd3.py#L1605
"""
if args.train_text_encoder:
(
transformer,
text_encoder_one,
text_encoder_two,
optimizer,
train_dataloader,
lr_scheduler,
) = accelerator.prepare(
transformer, text_encoder_one, text_encoder_two, optimizer, train_dataloader, lr_scheduler
)
assert text_encoder_one is not None
assert text_encoder_two is not None
assert text_encoder_three is not None
"""
why assert text_encoder_three is not None in SD3 dreambooth traning script? it seems only text_encoder amd text_encoder_two are used for training