Which parallelism technique is used in hugging face accelerate by default if we don't specify any technique in accelerate config but use multiple gpus

Question

Which parallelism technique is used in hugging face accelerate by default if we don't specify any technique in accelerate config but use multiple gpus??

Only want to know the default technique used is it data parallel or distributed data parallel

Model = accelerate.prepare(model)
Model = accelerate.device()