DPMPP 2m sde 1 sample per step rather than 2?

[kalkal11]

Just digging a bit further into the technically side of things and I've noticed that dpmpp 2m sde appears to be using a single sample per iteration - this is in contrast with how automatic1111 webui seems to operate.

Perhaps I'm misunderstanding the code and there are in fact 2 samples?

However if not, why is there such a dramatic different for this implementation of the sampler between comfy and auto?

[kalkal11]

Just talking to a friend about it, should the 2m variant not consistently be 2 samples per iteration?

https://arxiv.org/pdf/2211.01095.pdf

[comfyanonymous]

I'm pretty sure both of us use the exact same code from k diffusion for this sampler.