What should the input be for sigma from scipy.optimize.curve_fit for aymmetric uncertainty?

Question

I read the documentation of scipy.optimize.curve_fit, but sigma is still unclear for me.
I did understand what sigma is here used for, but I did not understand, how should I actually use it with asymmetric uncertainty.
Documentation: https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.curve_fit.html#scipy.optimize.curve_fit

Detailed description:

The documentation says:

If we define residuals as r = ydata - f(xdata, *popt), then the interpretation of sigma depends on its number of dimensions:
A 1-D sigma should contain values of standard deviations of errors in ydata. In this case, the optimized function is chisq = sum((r / sigma) ** 2).

The definition of residual is fine. I will use it too, but I am confused with the 1-D sigma.

What I understood is (not sure if it is correct):
A 1-D sigma is a M-length sequence. Since the ydata here is a length M array, the value in the first position from sigma (sigma[0]) is the standard deviation for the errors of the value in the first position from ydata (ydata[0]).

Now, the question is, what should I use for sigma[0] if I have asymmetric errors?
I mean, if I have asymmetric errors by ydata[0] (e.g. +0.2 and -0.1), then I should calculate the standard deviation from 0.2 and -0.1? Did I understand correctly, that those yerrors I got directly, should not be used as sigma here?