I have two conda environments both installing torch==2.1.2, but one with nvidia-cublas-cu12(use pip install torch==2.1.2 recently), the other doesn’t(i install it maybe one year ago):
The One without nvidia-cublas-cu12 can run properly, So I tried to Uninstall nvidia-cublas-cu12 for the one with it, but it raises error like:ValueError: libcublas.so.*[0-9] not found in the system path
I am wondering what’s the difference and the reason why it happens
@Wayne20-ops
I tried to use basic torch without nvidia-cublas, and i was able to do so
Are you trying to use cuda in your code?
when I just run “import torch”, it will raise error because of the cublas missing
I did not face that.
Can you provide the outputs of the below commands
pip list | grep cublas
pip list | grep torch
pip list | grep numpy
$ pip list | grep numpy
numpy 1.26.4
numpydoc 1.5.0
$ pip list | grep torch
torch 2.1.2
$ pip list | grep cublas
@Wayne20-ops try to uninstall numpy and install it back again. Mine shows 2.2.2
I tried to install numpy the same as yours, but still get the same error