Cannot use standard pytorch multi gpu feature

[Xinghui-Li]

Problem Minkowski Engine is unable to incorporate pytorch multi gpu features. I train the model using standard pytorch multi gpu feature: model = torch.nn.DataParallel(model)

and the error occur when using multi gpu:

File "/home/xinghui/miniconda3/envs/mink/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/xinghui/miniconda3/envs/mink/lib/python3.7/site-packages/MinkowskiEngine-0.5.4-py3.7-linux-x86_64.egg/MinkowskiEngine/MinkowskiConvolution.py", line 302, in forward
    assert input.D == self.dimension
  File "/home/xinghui/miniconda3/envs/mink/lib/python3.7/site-packages/torch/nn/modules/module.py", line 779, in __getattr__
    type(self).__name__, name))
torch.nn.modules.module.ModuleAttributeError: 'MinkowskiConvolution' object has no attribute 'dimension'

• Please complete all sections of this template if applicable. For installation, you must report the environment. Otherwise, your issue will be closed automatically.

---

To Reproduce

import torch
import MinkowskiEngine as ME

class Test(torch.nn.Module):
    def __init__(self):
        super(Test, self).__init__()
        nn_modules = list()
        nn_modules.append(
            ME.MinkowskiConvolution(1, 2, kernel_size=3, dimension=2))
        nn_modules.append(ME.MinkowskiConvolution(2, 2, kernel_size=3, dimension=2))
        self.conv = torch.nn.Sequential(*nn_modules)
    def forward(self, x):
        return self.conv(x)

model = Test()
model = model.cuda()
model = torch.nn.DataParallel(model)

data = [
    [0, 0, 2.1, 0, 0, 2],
    [0, 1, 1.4, 3, 0, 1],
    [0, 0, 4.0, 0, 0, 0.5],
    [1, 0, 0, 2, 1],
    [0, 0, 0, 0, 0]
]

def to_sparse_coo(data):
    # An intuitive way to extract coordinates and features
    coords, feats = [], []
    for i, row in enumerate(data):
        for j, val in enumerate(row):
            if val != 0:
                coords.append([i, j])
                feats.append([val])

    return torch.IntTensor(coords), torch.FloatTensor(feats)

coords, feats = to_sparse_coo(data)
coords, feats = ME.utils.sparse_collate(
    coords=[coords], feats=[feats])

x = ME.SparseTensor(features=feats.cuda(), coordinates=coords.cuda())
x = model(x)

---

Expected behavior The code should not encounter any problem

---

Desktop (please complete the following information):

• OS: [e.g. Ubuntu 18.04]
• Python version: [e.g. 3.7]
• Pytorch version: [e.g. 1.7.1]
• CUDA version: [e.g. 10.2]
• NVIDIA Driver version: [e.g. 450.11]
• Minkowski Engine version [e.g. 0.5.0]
• Output of the following command. (If you installed the latest MinkowskiEngine, paste the output of python -c "import MinkowskiEngine as ME; ME.print_diagnostics()". Otherwise, paste the output of the following command.)

wget -q https://raw.githubusercontent.com/NVIDIA/MinkowskiEngine/master/MinkowskiEngine/diagnostics.py ; python diagnostics.py

---

Additional context Add any other context about the problem here.

[capoommain]

Hi @Xinghui-Li , Any news about how to solve this error?

[yuvalH9]

I got similar issue with:

1. Ubuntu 18.04.6
2. Python 3.8.16
3. ME 0.5.4
4. Pytorch 0.13.1 +cu117
5. Cuda 11.7
6. GPU Quadro RTX 6000

Is there any new regarding it?