The maximum number of tokens to sample, as requested by the server. The client MAY choose to sample fewer tokens than requested.