I’m not sure I understand what you mean by setting the initial state of the delay line, so let me know if this isn’t answering your question.
Using stackables or mults, A out goes into A in, B out into B in. Self oscillation starts on channel A once you turn the gain past around +/- 1/3 of the way, and B is always at 1 gain (so using a compressor or VCA in between the feedback loop leads to more usable results. True for channel A as well, but the patch will still work without doing this). Then the delay time for each channel sets the pitch of each oscillator.
So, if I want to use A as carrier and B as modulator, I just tap B’s signal out into the modulation input for A, and then tap the signal out of A into my mixer or whatever. Or you can even feed B into its own modulation input and then into A, creating that classic feedback modulator algorithm.
EDIT: My bad, I totally forgot that you need to invert B’s feedback signal in order to get it to oscillate properly. So if it’s not giving you self-oscillation that might be why. Leaving it uninverted just produces noise or silence.