One model call, end to end

Follow one request from text to tokens to vectors to generated words, then make your first call.