num_epochs = 10 

-batch_size", type=int, default=1

--gradient_accumulation_steps = 4

--learning_rate = 2e-4

--lora_alpha", type=int, default=16

--merge_and_save", action="store_true