How does k-diffusion work with stable diffusion model exactly?

[shenxiul]

I was struggling to understand the codebase where we hooked up k-diffusion samplers with stable diffusion model. From stable diffusion repo, the corresponding model is expecting an input with a timestep, whereas k-diffusion seems to expect an input with a sigma, how exactly did we make it compatible? Looks like we did some kind of adaptor to make the model being able to accept sigma instead of timestep, but where is that exactly? Or I misunderstood the mechanism?