Token count of the embedded input (incl. the pooling token).
L2-normalized embedding (length = model hidden size).
Token count of the embedded input (incl. the pooling token).