DataLoader Parallelism fails

Hi. This will only use one core on my machine :
I’m using Gulpio to load the data. I don’t think the error comes from there though, after analyze of the differences between gulpio DataLoader and pytorch’s.

from multiprocessing import set_start_method
if __name__ === '__main__':
    #nothing makes it work properly

#no matter what value I put in there

dataset_train = GulpVideoDataset('/hdd/cropped_videos/gulped_train', \
            num_frames=num_frames, target_fps=target_fps, is_val=False)
data_loader_train = DataLoader(dataset_train, batch_size=bs, \
            sampler=WeightedRandomSampler(weights=weights,num_samples=dataset_length_train, \
            replacement=True), drop_last=True, num_workers=threads)

for data, labels in data_loader_train:

pip3 list gives

gulpio (521.36)
torch (0.3.0.post4)
torchvision (0.2.0)

Do you know what I’m doing wrong please ?