Erf implementation under the hood

I am trying to find how erf function is implemented in pytorch but cannot find a code pointer. I am assuming being an integral it must have been approximated but there is no references on how so.
I found this commit which contains erf inverse being approximated and then corrected using newton raphson, is there something similar being done for erf or does pytorch use existing implementations from std?

Yes, it seems the standard definition is used here.