he-zh / sparse-double-descent Goto Github PK

When trying to run the first example, the following error appears:
root@saa-l-pc:/home/user/sparse/sparse-double-descent# python main.py lottery --training_steps=160ep --lr=0.1 --milestone_steps=80ep,120ep --gamma=0.1 --rewinding_steps=1000it --de fault_hparams=cifar_pytorch_resnet_18 --levels=30 --random_labels_fraction=0.2 --dataset_name=cifar10 --gpu=0 --fix_all_random_seeds=1
Traceback (most recent call last):
File "/home/user/sparse/sparse-double-descent/main.py", line 13, in
from experiments import runner_registry
File "/home/user/sparse/sparse-double-descent/experiments/runner_registry.py", line 7, in
from training.runner import TrainingRunner
File "/home/user/sparse/sparse-double-descent/training/runner.py", line 9, in
from utils import shared_args
File "/home/user/sparse/sparse-double-descent/utils/shared_args.py", line 10, in
import models.registry
File "/home/user/sparse/sparse-double-descent/models/registry.py", line 12, in
from models import cifar_vgg, mnist_mlp, imagenet_resnet, cifar_pytorch_resnet, tinyimagenet_resnet
File "/home/user/sparse/sparse-double-descent/models/cifar_vgg.py", line 10, in
from pruning import magnitude
File "/home/user/sparse/sparse-double-descent/pruning/magnitude.py", line 10, in
import models.base
File "/home/user/sparse/sparse-double-descent/models/base.py", line 127, in
class DistributedDataParallel(Model, torch.nn.parallel.DistributedDataParallel):
File "/root/miniconda3/lib/python3.9/abc.py", line 106, in new
cls = super().new(mcls, name, bases, namespace, **kwargs)
TypeError: Cannot create a consistent method resolution
order (MRO) for bases ABC, Module

pytorch-1.12.1, torchvision-0.13.1, but the same with older packages.
Thank you in advance.