Because I used pretrained vectors of that size which were trained on Google News. If you train your own word embeddings you could experiment with different sizes for the dimension.