Numba code for comparison against http://jakevdp.github.com/blog/2012/08/08/memoryview-benchmarks/

numba_sweetness.py

from numba.decorators import jit as jit

@jit(arg_types=[[['d']], [['d']]], ret_type=[['d']])
def pairwise_numba(X, output):
    n_samples, n_dim = X.shape
    n_samples1, n_samples2 = output.shape
    
    for ii in range(n_samples):
        for jj in range(n_samples):
            result = 0.0;
            for kk in range(n_dim):
                result += (X[ii,kk] - X[jj,kk]) * (X[ii,kk] - X[jj,kk])
            output[ii,jj] = result
    
    return output

Just out of curiosity: why do you "import jit as jit"?