Get AttributeError: type object 'object' has no attribute 'SUM'

[hangfei]

Bug description

Getting this error:

self, tensor: Tensor, group: Optional[Any] = None, reduce_op: Optional[Union[ReduceOp, str]] = ReduceOp.SUM

E AttributeError: type object 'object' has no attribute 'SUM'

How to reproduce the bug

Install pyg and run pytest.

Error messages and logs

test/conftest.py:7: in <module>
    from torch_geometric.data import Dataset
torch_geometric/__init__.py:4: in <module>
    import torch_geometric.data
torch_geometric/data/__init__.py:7: in <module>
    from .lightning_datamodule import (
torch_geometric/data/lightning_datamodule.py:24: in <module>
    from pytorch_lightning import LightningDataModule as PLLightningDataModule
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/__init__.py:35: in <module>
    from pytorch_lightning.callbacks import Callback  # noqa: E402
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/callbacks/__init__.py:28: in <module>
    from pytorch_lightning.callbacks.pruning import ModelPruning
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/callbacks/pruning.py:31: in <module>
    from pytorch_lightning.core.module import LightningModule
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/core/__init__.py:16: in <module>
    from pytorch_lightning.core.module import LightningModule
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/core/module.py:47: in <module>
    from pytorch_lightning.trainer.connectors.logger_connector.fx_validator import _FxValidator
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/trainer/__init__.py:17: in <module>
    from pytorch_lightning.trainer.trainer import Trainer
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py:57: in <module>
    from pytorch_lightning.loops import PredictionLoop, TrainingEpochLoop
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/loops/__init__.py:15: in <module>
    from pytorch_lightning.loops.batch import TrainingBatchLoop  # noqa: F401
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/loops/batch/__init__.py:15: in <module>
    from pytorch_lightning.loops.batch.training_batch_loop import TrainingBatchLoop  # noqa: F401
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/loops/batch/training_batch_loop.py:20: in <module>
    from pytorch_lightning.loops.optimization.manual_loop import _OUTPUTS_TYPE as _MANUAL_LOOP_OUTPUTS_TYPE
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/loops/optimization/__init__.py:15: in <module>
    from pytorch_lightning.loops.optimization.manual_loop import ManualOptimization  # noqa: F401
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/loops/optimization/manual_loop.py:23: in <module>
    from pytorch_lightning.loops.utilities import _build_training_step_kwargs, _extract_hiddens
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/loops/utilities.py:29: in <module>
    from pytorch_lightning.strategies.parallel import ParallelStrategy
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/strategies/__init__.py:16: in <module>
    from pytorch_lightning.strategies.colossalai import ColossalAIStrategy  # noqa: F401
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/strategies/colossalai.py:48: in <module>
    class ColossalAIStrategy(DDPStrategy):
../../../opt/anaconda3/envs/sparse_tensor/lib/python3.8/site-packages/pytorch_lightning/strategies/colossalai.py:435: in ColossalAIStrategy
    self, tensor: Tensor, group: Optional[Any] = None, reduce_op: Optional[Union[ReduceOp, str]] = ReduceOp.SUM
E   AttributeError: type object 'object' has no attribute 'SUM'

Environment

#- Lightning Component (e.g. Trainer, LightningModule, LightningApp, LightningWork, LightningFlow):
#- PyTorch Lightning Version (e.g., 1.5.0):
#- Lightning App Version (e.g., 0.5.2):
#- PyTorch Version (e.g., 1.10):
#- Python version (e.g., 3.9):
#- OS (e.g., Linux):
#- CUDA/cuDNN version:
#- GPU models and configuration:
#- How you installed Lightning(`conda`, `pip`, source):
#- Running environment of LightningApp (e.g. local, cloud):

More info

No response

cc @tchaton @rohitgr7

[agoel00]

I am getting the same error on M1 Mac while importing pytorch_lightning

[hangfei]

For local testing, i was able to work around it by pip uninstall pytorch_lightning.

[adamjstewart]

Same error for me on M1:

• Lightning Component: all
• PyTorch Lightning Version: 1.8.0
• Lightning App Version: N/A
• PyTorch Version: 1.12.1
• Python version: 3.9.13
• OS: macOS 12.6.1 (arm64)
• CUDA/cuDNN version: N/A
• GPU models and configuration: N/A
• How you installed Lightning: spack
• Running environment of LightningApp: N/A

Downgrading to 1.7.7 solved the issue for now. Hoping this gets fixed in 1.8.1.

[agoel00]

Downgrading to 1.7.7 solved it for me as well. I think the issue is due to the ColossalAI Integration added in 1.8

[awaelchli]

This linked PR should fix this issue. Thanks for reporting! This bug affects all users with a torch installation where torch.distributed.is_available() returns False.

[adamjstewart]

This isn't the first time PyTorch Lightning broke support for PyTorch build without distributed support: #10348

I wonder if it would be possible to test PyTorch without distributed support in CI...

[awaelchli]

I have already investigated in that direction. One could change it in the CI like you said. For now, I pretend distributed is not available. What I have added so far is this regression test.