A weird problem that execute "import torch" in bash works but when you run it in Jupyter notebook:
ImportError Traceback (most recent call last) <ipython-input-4-8ba1970b60ce> in <module> 6 import random 7 ----> 8 import torch 9 import torch.nn as nn 10 ~/miniconda3/envs/tf/lib/python3.6/site-packages/torch/__init__.py in <module> 79 del _dl_flags 80 ---> 81 from torch._C import * 82 83 __all__ += [name for name in dir(_C) ImportError: dlopen: cannot load any more object with static TLS
Seems that it's a compatibility issue. Even though there are a large volume of discussions but none of them works. Ironically, I accidently fixed the issue by resolving another: # RuntimeError cuDNN error CUDNN_STATUS_EXECUTION_FAILED Solution
I marked these solutions as "potential" because I didn't make a strict control variable experiment. I tried all of them and it worked. :-)
1. Ubuntu 14.04 LTS 2. Python 3.6.9 3. PyTorch 1.3.0 4. cuda 10.1 (Tesla K40m)
The order of following solutions matters.
To the best of my understanding, this operation does the trick (but didn't verified). Downgrade cuda from 10.x to 9.0.
conda install pytorch==1.2.0 torchvision==0.4.0 cudatoolkit=10.0 -c pytorch
If you use conda, you can refer to this page to find a working combination: INSTALLING PREVIOUS VERSIONS OF PYTORCH
$ conda update jupyter
do-release-upgrade from 14.04 to 16.04 and finally
18.04 LTS. Ubuntu