My python implementation of pseudo-rng algorithm from http://amca01.wordpress.com/2012/07/02/a-conceptually-simple-prng/

prng.py

def prng(n):
    p = (1 << 31) - 1
    arr = [0 for i in range(n + 1)]
    arr[0] = 7
    arr[1] = 7

    for i in range(2, n + 1):
        arr[i] = pow_mod(7, arr[i - 1], p) + pow_mod(7, arr[i - 2], p)

    return arr

def pow_mod(base, power, p):
    if power == 0:
        return 1
    result = base
    for k in range(2, power + 1):
        result *= base  # result == base ** k
        if result >= p:
            c = result % p
            # d - how many times we multiple (base ** k) * (base ** k) * ...
            # e - the rest of power
            d, e = divmod(power, k)
            # now pow_mod(base, power, p) ==
            #     pow_mod(c, d, p)     -- fast calculation using mod properties
            #   * pow_mod(base, e, p)  -- the rest
            #
            # mod properties explained:
            # (base ** k) * (base ** k) * ... [d times] ... * (base ** k) mod p
            # == ((base ** k) mod p) ** d
            return (pow_mod(c, d, p) * pow_mod(base, e, p)) % p
    else:
        return result

def test_pow_mod():
    assert pow_mod(7, 100, 9999) == 7 ** 100 % 9999

test_pow_mod()

from pprint import pprint
pprint(prng(50))

It generates 1 million numbers in 13 seconds when launched under pypy 1.9