ImportError: libcupti.so.11.2: cannot open shared object file: No such file or directory

[marcuswang6]

🐛 Describe the bug

---

ImportError Traceback (most recent call last)
4 import random
5 import copy
----> 6 import torch
7 #import learn2learn as l2l
8 import numpy as np

File /dev_data_1/conda/envs/meta/lib/python3.8/site-packages/torch/init.py:202, in
200 if USE_GLOBAL_DEPS:
201 _load_global_deps()
--> 202 from torch._C import * # noqa: F403
204 # Appease the type checker; ordinarily this binding is inserted by the
205 # torch._C module initialization code in C
206 if TYPE_CHECKING:

ImportError: libcupti.so.11.2: cannot open shared object file: No such file or directory

Versions

torch 1.12.0
python 3.8.13

cc @malfet @seemethere @ngimel

[marcuswang6]

I have tried below:

1. sudo ldconfig /usr/local/cuda-11.7/lib64
2. sudo vim ~/.bashrc and add:

export PATH=/usr/local/cuda-11.7/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.7/lib64:$LD_LIBRARY_PATH
export CUDA_HOME=/usr/local/cuda-11.7

3. find /usr/local/cuda/ -name libcupti.so.11.2 find the file.

[ngimel]

How are you building pytorch?

[RobinY99]

I have met a similar issue, here is my solution.
Environment: Ubuntu + Conda
Missing File: libcupti.so.12
Actually, I think it is a missing file error. So putting the required file to the correct location could solve that.

Solution:
Firstly, use command "sudo find / -name libcupti.so.11.2". (You might find a required file existing in your system).
Secondly, use command "cp $your_file_path$ ~/.conda/envs/$your_env_name$/lib/python_your_version/site-packages/nvidia/nccl/lib"
I believe you could the where the file is and the location for it.

[MiangChen]

Ubuntu 22.04(Kernel 6.8.0) + Anaconda
NVIDIA Driver 535.216.01 + CUDA 12.2 + CuDNN 9.8.0
Python 3.10 torch==2.5.1+cu124

This is my solution:

• find libcupti.so.12:

sudo find / -name libcupti.so.12

There may be many same files, choose the one in the folder cuda-xx.x:

• create a folder in your conda environment:

mkdir -p ~/anaconda3/envs/{your_env_name}/lib/{your_python_version}/site-packages/nvidia/nccl/lib

• make link for libcupti.so.12:

ln -s /usr/local/cuda-12.2/extras/CUPTI/lib64/libcupti.so.12 ~/anaconda3/envs/{your_env_name}/lib/{your_python_version}/site-packages/nvidia/nccl/lib/libcupti.so.12

[WarrenK17]

I suddenly met the same problem,although I could use the environment hours ago.
I just copy the site-packages/nvidia in usable envirenment into the broken environment,which is just empty.