Does TokenEmbedding size must equal to embed size * vocab size in Transformer model

I have long sequences and the vocab is huge resulting to OOM