IFFTN imaginary part

I am testing with ifftn, and I am seeing an imaginary part in the value, which I think does not make sense.
In here, we see that we don’t have an imaginary part (it mentions that it ignores it)
, but in the case of ifftn, even in pytorch own example, we have imaginary parts (they are small though).
Should I just ignore them, and use the real values? And what would happen if I use fft and ifft multiple times? Do I lose any information? fft and ifft should neither add or lose any information.
Could someone please elaborate this for me?