In LoRA training, Alpha functions as a scaling factor; the effective learning rate approximates Alpha/rank, and when the two are equal the scaling ratio is 1.