Curiousity of seq2seq model

After reading the seq2seq google paper it was stated that they used a fixed vector length of 160000 and 80000 for source and target sequence respectively.
I’m wondering how google vectorized their data, whether as one-hot encoding ([0,1,0,0,0…0]) or as zero padded vector of unique token index sequence ([10, 17, 2, 3, 0, 0 …0, 0, 0]) where each interger > 0 represents an index of word in the vocabulary.
Can someone please throw more light on this.