torch_sparse/_convert_cpu.so: undefined symbol: _ZNK2at6Tensor5zero_Ev

[shuimei]

code:
In [1]: import torch
In [2]: from torch_geometric.data import Data

error:

OSError Traceback (most recent call last)
in ()
----> 1 from torch_geometric.data import Data

~/anaconda3/lib/python3.7/site-packages/torch_geometric/init.py in ()
5 from .debug import is_debug_enabled, debug, set_debug
6
----> 7 import torch_geometric.data
8 import torch_geometric.loader
9 import torch_geometric.transforms

~/anaconda3/lib/python3.7/site-packages/torch_geometric/data/init.py in ()
----> 1 from .data import Data
2 from .hetero_data import HeteroData
3 from .temporal import TemporalData
4 from .batch import Batch
5 from .dataset import Dataset

~/anaconda3/lib/python3.7/site-packages/torch_geometric/data/data.py in ()
1 from typing import (Optional, Dict, Any, Union, List, Iterable, Tuple,
2 NamedTuple, Callable)
----> 3 from torch_geometric.typing import OptTensor, NodeType, EdgeType
4 from torch_geometric.deprecation import deprecated
5

~/anaconda3/lib/python3.7/site-packages/torch_geometric/typing.py in ()
2
3 from torch import Tensor
----> 4 from torch_sparse import SparseTensor
5
6 # Types for accessing data ####################################################

~/anaconda3/lib/python3.7/site-packages/torch_sparse/init.py in ()
14 ]:
15 torch.ops.load_library(importlib.machinery.PathFinder().find_spec(
---> 16 f'{library}_{suffix}', [osp.dirname(file)]).origin)
17
18 if torch.cuda.is_available(): # pragma: no cover

~/anaconda3/lib/python3.7/site-packages/torch/_ops.py in load_library(self, path)
108 # static (global) initialization code in order to register custom
109 # operators with the JIT.
--> 110 ctypes.CDLL(path)
111 self.loaded_libraries.add(path)
112

~/anaconda3/lib/python3.7/ctypes/init.py in init(self, name, mode, handle, use_errno, use_last_error)
354
355 if handle is None:
--> 356 self._handle = _dlopen(self._name, mode)
357 else:
358 self._handle = handle

OSError: /root/anaconda3/lib/python3.7/site-packages/torch_sparse/_convert_cpu.so: undefined symbol: _ZNK2at6Tensor5zero_Ev

I followed the Quick Start installation, with torch=1.10.0+cu102

[rusty1s]

This is related to a potentially false intallation of torch-sparse. Can you show me the installation log when running:

pip uninstall torch-sparse
pip uninstall torch-sparse
pip install torch-sparse -f https://data.pyg.org/whl/torch-1.10.0+cu102.html

[guillaumehu]

There is the same issue when running the Colab Notebooks tutorials.

[rusty1s]

Colab notebooks are fixable as suggested here: pyg-team/pytorch_geometric#3484

[guillaumehu]

Great, thanks a lot !

[dreamcoils]

This is related to a potentially false intallation of torch-sparse. Can you show me the installation log when running:

pip uninstall torch-sparse
pip uninstall torch-sparse
pip install torch-sparse -f https://data.pyg.org/whl/torch-1.10.0+cu102.html

it works!!
thanks a lot!

[shuimei]

Found existing installation: torch-sparse 0.6.12 Uninstalling torch-sparse-0.6.12: Would remove: /root/anaconda3/lib/python3.7/site-packages/test/* /root/anaconda3/lib/python3.7/site-packages/torch_sparse-0.6.12.dist-info/* /root/anaconda3/lib/python3.7/site-packages/torch_sparse/* Would not remove (might be manually added): /root/anaconda3/lib/python3.7/site-packages/test/test_basis.py /root/anaconda3/lib/python3.7/site-packages/test/test_broadcasting.py /root/anaconda3/lib/python3.7/site-packages/test/test_conv.py /root/anaconda3/lib/python3.7/site-packages/test/test_fps.py /root/anaconda3/lib/python3.7/site-packages/test/test_gather.py /root/anaconda3/lib/python3.7/site-packages/test/test_graclus.py /root/anaconda3/lib/python3.7/site-packages/test/test_grid.py /root/anaconda3/lib/python3.7/site-packages/test/test_knn.py /root/anaconda3/lib/python3.7/site-packages/test/test_multi_gpu.py /root/anaconda3/lib/python3.7/site-packages/test/test_nearest.py /root/anaconda3/lib/python3.7/site-packages/test/test_radius.py /root/anaconda3/lib/python3.7/site-packages/test/test_rw.py /root/anaconda3/lib/python3.7/site-packages/test/test_sampler.py /root/anaconda3/lib/python3.7/site-packages/test/test_scatter.py /root/anaconda3/lib/python3.7/site-packages/test/test_segment.py /root/anaconda3/lib/python3.7/site-packages/test/test_weighting.py /root/anaconda3/lib/python3.7/site-packages/test/test_zero_tensors.py Proceed (Y/n)? Successfully uninstalled torch-sparse-0.6.12 ~

pip install torch-sparse -f https://data.pyg.org/whl/torch-1.10.0+cu102.html Looking in links: https://data.pyg.org/whl/torch-1.10.0+cu102.html Collecting torch-sparse Downloading https://data.pyg.org/whl/torch-1.10.0%2Bcu102/torch_sparse-0.6.12-cp37-cp37m-linux_x86_64.whl (2.9 MB) |████████████████████████████████| 2.9 MB 500 kB/s Requirement already satisfied: scipy in ./anaconda3/lib/python3.7/site-packages (from torch-sparse) (1.1.0) Installing collected packages: torch-sparse

but the problem remains
now another OSError occurs:
OSError: /root/anaconda3/lib/python3.7/site-packages/torch_scatter/_scatter_cpu.so: undefined symbol: _ZNK2at6Tensor6narrowElll

my torch version is 1.10.0+cu102

[rusty1s]

You also need to uninstall and update torch-scatter, similarly to how you did it with torch-sparse.

[github-actions]

This issue had no activity for 6 months. It will be closed in 2 weeks unless there is some new activity. Is this issue already resolved?

[jingyanliao]

I'm trying to solve the same problem under torch 1.10.0+cu111, but after re-installing torch-scatter and torch-sparse the problem is still unsolved.

command I used:
pip uninstall torch-scatter pip uninstall torch-sparse pip install torch-sparse -f https://data.pyg.org/whl/torch-1.10.0+cu111.html pip install torch-scatter -f https://data.pyg.org/whl/torch-1.10.0+cu111.html

[rusty1s]

Can you install via

pip install torch-scatter==2.0.9 -f https://data.pyg.org/whl/torch-1.10.0+cu111.html

[Sparrow711]

Hey! I am having the same problem. And I tried to resolve this issue using the solution mentioned in this thread. Following is the error that I receive:

/home/anshal/anaconda3/envs/pointcept/lib/python3.10/site-packages/torch_geometric/typing.py:31: UserWarning: An issue occurred while importing 'torch-scatter'. Disabling its usage. Stacktrace: /home/anshal/anaconda3/envs/pointcept/lib/python3.10/site-packages/torch_scatter/_scatter_cuda.so: undefined symbol: _ZNK3c107SymBool10guard_boolEPKcl
warnings.warn(f"An issue occurred while importing 'torch-scatter'. "
Traceback (most recent call last):
File "/home/anshal/summer_intern/Pointcept-main/exp/scannet/semseg-pt-v2m2-0-base/code/tools/train.py", line 9, in
from pointcept.engines.train import Trainer
File "/home/anshal/summer_intern/Pointcept-main/exp/scannet/semseg-pt-v2m2-0-base/code/pointcept/engines/train.py", line 25, in
from pointcept.models import build_model
File "/home/anshal/summer_intern/Pointcept-main/exp/scannet/semseg-pt-v2m2-0-base/code/pointcept/models/init.py", line 7, in
from .point_transformer_v2 import *
File "/home/anshal/summer_intern/Pointcept-main/exp/scannet/semseg-pt-v2m2-0-base/code/pointcept/models/point_transformer_v2/init.py", line 8, in
from .point_transformer_v2m1_origin import *
File "/home/anshal/summer_intern/Pointcept-main/exp/scannet/semseg-pt-v2m2-0-base/code/pointcept/models/point_transformer_v2/point_transformer_v2m1_origin.py", line 14, in
from torch_scatter import segment_csr
File "/home/anshal/anaconda3/envs/pointcept/lib/python3.10/site-packages/torch_scatter/init.py", line 16, in
torch.ops.load_library(spec.origin)
File "/home/anshal/.local/lib/python3.10/site-packages/torch/_ops.py", line 573, in load_library
ctypes.CDLL(path)
File "/home/anshal/anaconda3/envs/pointcept/lib/python3.10/ctypes/init.py", line 374, in init
self._handle = _dlopen(self._name, mode)
OSError: /home/anshal/anaconda3/envs/pointcept/lib/python3.10/site-packages/torch_scatter/_scatter_cuda.so: undefined symbol: _ZNK3c107SymBool10guard_boolEPKcl
/home/anshal/anaconda3/envs/pointcept/lib/python3.10/site-packages/torch_geometric/typing.py:31: UserWarning: An issue occurred while importing 'torch-scatter'. Disabling its usage. Stacktrace: /home/anshal/anaconda3/envs/pointcept/lib/python3.10/site-packages/torch_scatter/_scatter_cuda.so: undefined symbol: _ZNK3c107SymBool10guard_boolEPKcl
warnings.warn(f"An issue occurred while importing 'torch-scatter'. "
Traceback (most recent call last):
File "/home/anshal/summer_intern/Pointcept-main/tools/train.py", line 9, in
from pointcept.engines.train import Trainer
File "/home/anshal/summer_intern/Pointcept-main/pointcept/engines/train.py", line 25, in
from pointcept.models import build_model
File "/home/anshal/summer_intern/Pointcept-main/pointcept/models/init.py", line 7, in
from .point_transformer_v2 import *
File "/home/anshal/summer_intern/Pointcept-main/pointcept/models/point_transformer_v2/init.py", line 8, in
from .point_transformer_v2m1_origin import *
File "/home/anshal/summer_intern/Pointcept-main/pointcept/models/point_transformer_v2/point_transformer_v2m1_origin.py", line 14, in
from torch_scatter import segment_csr
File "/home/anshal/anaconda3/envs/pointcept/lib/python3.10/site-packages/torch_scatter/init.py", line 16, in
torch.ops.load_library(spec.origin)
File "/home/anshal/.local/lib/python3.10/site-packages/torch/_ops.py", line 573, in load_library
ctypes.CDLL(path)
File "/home/anshal/anaconda3/envs/pointcept/lib/python3.10/ctypes/init.py", line 374, in init
self._handle = _dlopen(self._name, mode)
OSError: /home/anshal/anaconda3/envs/pointcept/lib/python3.10/site-packages/torch_scatter/_scatter_cuda.so: undefined symbol: _ZNK3c107SymBool10guard_boolEPKcl

I installed pytorch version 2.0.1, cuda 11.7 and torch-scatter 2.1.1

[Sruthik22]

I also have the same issue: OSError: .conda/envs/GIGN/lib/python3.9/site-packages/torch_sparse/_version_cpu.so: undefined symbol: _ZN5torch3jit17parseSchemaOrNameERKSs

[rusty1s]

This usually indicates a version conflict, so something with your CUDA and PyTorch version does not match with the installed version coming from torch-scatter. Any additional information about your system and how you installed torch-scatter would let me help you better.

[Sparrow711]

I have cuda 11.7 installed on my laptop. And in the conda environment I have pytorch 2.0.1 for cuda 11.7, torch-scatter (2.1.1+pt20cu117). And I used the following command to install pytorch-scatter, pytorch-sparse and pytorch cluster:
conda install pytorch-cluster pytorch-scatter pytorch-sparse -c pyg -y

[rusty1s]

How did you install PyTorch? It is preferred that you install it from the official binaries (not from something like conda-forge).

[Sparrow711]

I installed Pytorch using the instructions on their site. I used the following:
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia

[rusty1s]

Thanks for sharing. Can you try to run install torch-scatter (and so on) via pip (with -f option) rather than via conda? Note that you may need to uninstall the conda versions of it first. This should help me to track down the issue better.

[rookiie1918]

OSError: /home/lijuncai/app/anaconda3/envs/ljc/lib/python3.9/site-packages/torch_cluster/_grid_cuda.so: undefined symbol: _ZNK3c1010TensorImpl36is_contiguous_nondefault_policy_implENS_12MemoryFormatE
torch_version is 1.12.1+cu102

[mainguyenanhvu]

I found the solution here.

[viettham1998]

I also have the same issue:
OSError: /home/hvtham/.local/lib/python3.10/site-packages/torch_sparse/_convert_cpu.so: undefined symbol: _ZN3c1013ParallelGuardC1Eb

My torch_version is 2.2.1+cu121