weird results after calculating n_grammes perplexity

Question

I hope you will help me with solving this problem; so I trained 3 n_grammes models (n=1,2,3) but I found out weird results after calculating perple