How to pass arguments to worker_init_fn in multiprocessing dataloader?

1

I want to pass arguments to worker_init_fn in multiprocessing dataloader.

e.g.

def worker_init_fn(worker_id, **kwargs):
    pass

BUT in pytorch 1.3 only support one argument that is worker_id

I tried to use lambda or local function which don’t work with multiprocessing.

Any one can help me? Thanks

1 Like