libtorch_cpu.so：未定义符号：iJIT_IsProfilingActive

Question

我创建了这个Conda环境：

conda env create -f environment.yml

environment.yml

文件：

name: deep3d_pytorch
channels:
  - pytorch
  - conda-forge
  - defaults
dependencies:
  - python=3.6
  - pytorch=1.6.0
  - torchvision=0.7.0
  - numpy=1.18.1
  - scikit-image=0.16.2
  - scipy=1.4.1
  - pillow=6.2.1
  - pip
  - ipython=7.13.0
  - yaml=0.1.7
  - pip:
    - matplotlib==2.2.5
    - opencv-python==3.4.9.33
    - tensorboard==1.15.0
    - tensorflow==1.15.0
    - kornia==0.5.5
    - dominate==2.6.0
    - trimesh==3.9.20

我激活Conda环境。但即使是像

python -c "import torch; print(torch.__version__)"

这样获取 PyTorch 版本的简单语句也会引发

undefined symbol

错误：

(deep3d_pytorch) m3@i7:~/repos/Deep3DFaceRecon_pytorch> python -c "import torch; print(torch.__version__)"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/home/m3/anaconda3/envs/deep3d_pytorch/lib/python3.6/site-packages/torch/__init__.py", line 189, in <module>
    from torch._C import *
ImportError: /home/m3/anaconda3/envs/deep3d_pytorch/lib/python3.6/site-packages/torch/lib/libtorch_cpu.so: undefined symbol: iJIT_IsProfilingActive

我相信 Conda 安装的 PyTorch 已损坏。但Conda日志一切都很好。有人有线索或提示吗？我在本地计算机和 Google Colab 上都收到

undefined symbol

错误。

更新：最小环境

即使是像下面这样的最小环境，也会引发类似的错误：

conda create -n minimal_pytorch python=3.6 pytorch torchvision torchaudio -c pytorch
source activate minimal_pytorch && python -c "import torch; print(torch.__version__)"

抛出类似的

undefined symbol

错误：

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/usr/local/envs/minimal_pytorch/lib/python3.6/site-packages/torch/__init__.py", line 197, in <module>
    from torch._C import *  # noqa: F403
ImportError: /usr/local/envs/minimal_pytorch/lib/python3.6/site-packages/torch/lib/libtorch_cpu.so: undefined symbol: iJIT_NotifyEvent

Python版本

创建环境时省略Python版本时：

conda create -n minimal_pytorch python pytorch torchvision torchaudio -c pytorch

错误已解决：

source activate minimal_pytorch && python -c "import torch; print(torch.__version__)"

PyTorch 版本已收到，没有任何错误：

2.2.2

Answer 1

将 mkl 包降级到早期版本应该可以工作