Overflow encountered in long_scalars while generating large sparse matrix in scipy

I am using scipy rand/random to create a m x n sized sparse array for a Neural Network experiment. I understand from documentation that sparse matrices should work for NxN where N~10^7. However the program runs into a long_scalar overflow which results in a value error inside the rand function. at num_rows_arr[i_in] ~ 16000 and dim_w ~ 130000 the rand function gives this error