Trojan Detection Challenge - Evasive Trojans Track Forum

Go back to competition Back to thread list Post in this thread

> Submission Returns Errors

Our submission returns the following runtime errors. It's related to CUDA multiprocessing. Can we check the submission server runs correctly?

Traceback (most recent call last):
  File "/tmp/codalab/tmprWT7ov/run/program/evaluation.py", line 567, in <module>
    auroc1 = run_mntd_crossval(trojan_model_dir, clean_model_dir, num_folds=5)
  File "/tmp/codalab/tmprWT7ov/run/program/evaluation.py", line 466, in run_mntd_crossval
    train_meta_network(meta_network, train_loader)
  File "/tmp/codalab/tmprWT7ov/run/program/evaluation.py", line 396, in train_meta_network
    for i, (net, label) in enumerate(train_loader):
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/dataloader.py", line 521, in __next__
    data = self._next_data()
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/dataloader.py", line 1203, in _next_data
    return self._process_data(data)
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/dataloader.py", line 1229, in _process_data
    data.reraise()
  File "/usr/local/lib/python3.6/dist-packages/torch/_utils.py", line 434, in reraise
    raise exception
RuntimeError: Caught RuntimeError in DataLoader worker process 0.
Original Traceback (most recent call last):
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/_utils/worker.py", line 287, in _worker_loop
    data = fetcher.fetch(index)
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/_utils/fetch.py", line 49, in fetch
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/_utils/fetch.py", line 49, in <listcomp>
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/usr/local/lib/python3.6/dist-packages/torch/utils/data/dataset.py", line 363, in __getitem__
    return self.dataset[self.indices[idx]]
  File "/tmp/codalab/tmprWT7ov/run/program/evaluation.py", line 363, in __getitem__
    return torch.load(os.path.join(self.model_paths[index], 'model.pt')), self.labels[index]
  File "/usr/local/lib/python3.6/dist-packages/torch/serialization.py", line 607, in load
    return _load(opened_zipfile, map_location, pickle_module, **pickle_load_args)
  File "/usr/local/lib/python3.6/dist-packages/torch/serialization.py", line 882, in _load
    result = unpickler.load()
  File "/usr/local/lib/python3.6/dist-packages/torch/serialization.py", line 857, in persistent_load
    load_tensor(data_type, size, key, _maybe_decode_ascii(location))
  File "/usr/local/lib/python3.6/dist-packages/torch/serialization.py", line 846, in load_tensor
    loaded_storages[key] = restore_location(storage, location)
  File "/usr/local/lib/python3.6/dist-packages/torch/serialization.py", line 175, in default_restore_location
    result = fn(storage, location)
  File "/usr/local/lib/python3.6/dist-packages/torch/serialization.py", line 157, in _cuda_deserialize
    return obj.cuda(device)
  File "/usr/local/lib/python3.6/dist-packages/torch/_utils.py", line 71, in _cuda
    with torch.cuda.device(device):
  File "/usr/local/lib/python3.6/dist-packages/torch/cuda/__init__.py", line 272, in __enter__
    self.prev_idx = torch.cuda.current_device()
  File "/usr/local/lib/python3.6/dist-packages/torch/cuda/__init__.py", line 479, in current_device
    _lazy_init()
  File "/usr/local/lib/python3.6/dist-packages/torch/cuda/__init__.py", line 205, in _lazy_init
    "Cannot re-initialize CUDA in forked subprocess. To use CUDA with "
RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method

Posted by: neuronoverflow @ July 28, 2022, 9:56 p.m.

Hello,

Sorry for the late reply. The problem was that networks were loading directly onto the GPU, which was causing errors in the loader. This error is fixed now. You should be able to resubmit the same zip file successfully.

All the best,
Mantas (TDC co-organizer)

Posted by: mmazeika @ Aug. 2, 2022, 1:47 a.m.
Post in this thread