BabyLM Challenge: Exploring the Effect of Variation Sets on Language Model Training Efficiency