##Although this trial issue does not effect the normal running##

Definition
# Parameters that control generation
generation_batch_size: Optional[int] = field(
default=None,
metadata={
"help": "Batch size to use for generation. If None, it defaults to the effective training batch size: "
"per_device_train_batch_size * num_processes * gradient_accumulation_steps."
},
)

post_init

    if self.generation_batch_size is None:
        self.generation_batch_size = self.per_device_train_batch_size * num_processes *self.steps_per_generation

only steps_per_generation is not configured, gradient_accumulation_steps equals steps_per_generation