Colab PyTorch | ImportError: /usr/local/lib/python3.7/dist-packages/_XLAC.cpython-37m-x86_64-linux-gnu.so

Question

在 Google Colaboratory 上，我尝试了所有 3 种运行时：CPU、GPU、TPU。都报同样的错误。

单元格：

# NB: Only run in TPU environment
!pip install cloud-tpu-client==0.10 https://storage.googleapis.com/tpu-pytorch/wheels/torch_xla-1.8-cp37-cp37m-linux_x86_64.whl

!pip -q install pytorch-lightning==1.2.7 transformers torchmetrics awscli mlflow boto3 pycm

import os
import sys
import logging

from pytorch_lightning import LightningDataModule

错误：

ImportError                               Traceback (most recent call last)
<ipython-input-6-09509a67016b> in <module>()
      3 import logging
      4 
----> 5 from pytorch_lightning import LightningDataModule
      6 from torch.utils.data import DataLoader, Dataset
      7 from transformers import AutoTokenizer

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/__init__.py in <module>()
     26 _PROJECT_ROOT = os.path.dirname(_PACKAGE_ROOT)
     27 
---> 28 from pytorch_lightning import metrics  # noqa: E402
     29 from pytorch_lightning.callbacks import Callback  # noqa: E402
     30 from pytorch_lightning.core import LightningDataModule, LightningModule  # noqa: E402

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/metrics/__init__.py in <module>()
     12 # See the License for the specific language governing permissions and
     13 # limitations under the License.
---> 14 from pytorch_lightning.metrics.classification import (  # noqa: F401
     15     Accuracy,
     16     AUC,

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/metrics/classification/__init__.py in <module>()
     12 # See the License for the specific language governing permissions and
     13 # limitations under the License.
---> 14 from pytorch_lightning.metrics.classification.accuracy import Accuracy  # noqa: F401
     15 from pytorch_lightning.metrics.classification.auc import AUC  # noqa: F401
     16 from pytorch_lightning.metrics.classification.auroc import AUROC  # noqa: F401

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/metrics/classification/accuracy.py in <module>()
     16 import torch
     17 
---> 18 from pytorch_lightning.metrics.functional.accuracy import _accuracy_compute, _accuracy_update
     19 from pytorch_lightning.metrics.metric import Metric
     20 

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/metrics/functional/__init__.py in <module>()
     12 # See the License for the specific language governing permissions and
     13 # limitations under the License.
---> 14 from pytorch_lightning.metrics.functional.accuracy import accuracy  # noqa: F401
     15 from pytorch_lightning.metrics.functional.auc import auc  # noqa: F401
     16 from pytorch_lightning.metrics.functional.auroc import auroc  # noqa: F401

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/metrics/functional/accuracy.py in <module>()
     16 import torch
     17 
---> 18 from pytorch_lightning.metrics.classification.helpers import _input_format_classification, DataType
     19 
     20 

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/metrics/classification/helpers.py in <module>()
     17 import torch
     18 
---> 19 from pytorch_lightning.metrics.utils import select_topk, to_onehot
     20 from pytorch_lightning.utilities import LightningEnum
     21 

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/metrics/utils.py in <module>()
     16 import torch
     17 
---> 18 from pytorch_lightning.utilities import rank_zero_warn
     19 
     20 METRIC_EPS = 1e-6

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/utilities/__init__.py in <module>()
     46 )
     47 from pytorch_lightning.utilities.parsing import AttributeDict, flatten_dict, is_picklable  # noqa: F401
---> 48 from pytorch_lightning.utilities.xla_device import XLADeviceUtils  # noqa: F401
     49 
     50 _TPU_AVAILABLE = XLADeviceUtils.tpu_device_exists()

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/utilities/xla_device.py in <module>()
     21 
     22 if _XLA_AVAILABLE:
---> 23     import torch_xla.core.xla_model as xm
     24 
     25 #: define waiting time got checking TPU available in sec

/usr/local/lib/python3.7/dist-packages/torch_xla/__init__.py in <module>()
    126 import torch
    127 from ._patched_functions import _apply_patches
--> 128 import _XLAC
    129 
    130 

ImportError: /usr/local/lib/python3.7/dist-packages/_XLAC.cpython-37m-x86_64-linux-gnu.so: undefined symbol: _ZN2at11result_typeERKNS_6TensorEN3c106ScalarE

Answer 1

在线搜索；同样的问题似乎有很多原因。

在我的例子中，在 Google Colaboratory 中将 Accelerator 设置为 None 解决了这个问题。

Answer 2

似乎 Google Colab 在安装 cloud-tpu-client 时与使用 GPU 混淆。我卸载了这个包，然后它就正常工作了。

!pip uninstall cloud-tpu-client

Answer 3

当torch和torch-xla版本不同时会出现这个错误。您可以使用 pip list | grep torch

检查两者的版本

Colab PyTorch | ImportError: /usr/local/lib/python3.7/dist-packages/_XLAC.cpython-37m-x86_64-linux-gnu.so

Colab PyTorch | ImportError: /usr/local/lib/python3.7/dist-packages/_XLAC.cpython-37m-x86_64-linux-gnu.so

python

pytorch

google-colaboratory

pytorch-lightning