how to set sampler when resume training?

[TinaChen95]

I'm training on a big dataset, and some times training may got interrupted before one epoch is done.
when I resume training, sampler.set_epoch seems to restart from step 0
how can I set the sampler to skip the first n steps to match my training ?