Open TinaChen95 opened 1 month ago
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 ?
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 ?