Is it possible that numba jit slows down my gcd execution?

I'm trying to compute a large amount of GCD(x,y) as part of Euler 625 and since this takes a lot of time, I tried adding @numba jit to speed it up.

I checked similar question like this, this and this but they are not helpful (I'm not using numpy or using a parallel-able functions (like sum) and I called the functions several times - so compiling laziness isn't the issue either) The results are :

with    numba: 8.220806121826172 seconds
without numba: 1.242861270904541 seconds

my code is added below.

1. why is numba slowing me down?

2. how can I execute this faster (the original limit is 10**11, which is incomputable in this way as it seems)

   import time
   from numba import jit, int32
   
   gcd_dict = {}
   
   #@jit
   def gcd(x, y ,k = None, reverse_k = None):
   
       if not k:
           k = (x,y)
           reverse_k = (y,x)
   
       if k in gcd_dict:
           return gcd_dict[k]
       elif reverse_k in gcd_dict:
           return gcd_dict[reverse_k]
   
       if not y:
           if len(gcd_dict) < 10**6:
               gcd_dict[k] = x
               gcd_dict[reverse_k] = x
           return x
       else:
           return gcd(y, x % y, k, reverse_k)
   
   
   
   def main():
       t = time.time()
       s = 0
   #    i_limit = 10**11+1
       i_limit = 10**3
   
       for i in range(1,i_limit):
           for j in range(1,i+1):
               s += gcd(i,j)
   
       print(s)
       print(time.time() - t)
   
   main()