AssertionError: Torch not compiled with CUDA enabled

[Testin1234]

I'm trying to do neural style swapping, and for some reason, I keep getting the following errors.
AssertionError: Torch not compiled with CUDA enabled
File "c:\apps\Miniconda3\lib\site-packages\torch\nn\modules\module.py", line 260, in cuda
return self._apply(lambda t: t.cuda(device))
File "c:\apps\Miniconda3\lib\site-packages\torch\nn\modules\module.py", line 187, in _apply
module._apply(fn)
File "c:\apps\Miniconda3\lib\site-packages\torch\nn\modules\module.py", line 187, in _apply
module._apply(fn)
File "c:\apps\Miniconda3\lib\site-packages\torch\nn\modules\module.py", line 193, in _apply
param.data = fn(param.data)
File "c:\apps\Miniconda3\lib\site-packages\torch\nn\modules\module.py", line 260, in
return self.apply(lambda t: t.cuda(device))
File "c:\apps\Miniconda3\lib\site-packages\torch\cuda_init.py", line 161, in _lazy_init
check_driver()
File "c:\apps\Miniconda3\lib\site-packages\torch\cuda_init.py", line 75, in _check_driver
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

I recently reinstalled conda and this has just been completely broken for me.

cc @ezyang

[zou3519]

How did you install pytorch? It sounds like you installed pytorch without CUDA support. https://pytorch.org/ has instructions for how to install pytorch with cuda support.

[Gailsunset]

Shalom!

I am new to pytorch, and already installed pytorch in my Mac laptop.

trying to run the wsi_bert.py code, and get this error.
"""
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled
"""
Please help.
Thx a lot!

[swj0418]

Try this.
conda install -c pytorch torchvision cudatoolkit=10.1 pytorch
Depending on what cuda version you have.

[ezyang]

Note that unless you know otherwise, your Mac probably doesn't have an NVIDIA GPU on it.

For further support questions related to installation, please follow up on the forums https://discuss.pytorch.org/

[AchinthyaBhat]

I don't know where to raise an issue, but this continues to be a problem on Windows when Pytorch is installed without CUDA support even though it has been recently fixed for MacOS. Can someone help me?

[dzhoshkun]

Experiencing this problem on Windows as well.

[jpryne]

I tried re-installing pytorch as per @swj0418 above,
conda install -c pytorch torchvision cudatoolkit=10.1 pytorch
yet to no avail. Running this from with a Jupyter notebook under Anaconda on Windows 10 with an Nvidia1060 6GB & Intel i7.

[YuanYingtao]

Experiencing this problem on Windows as well.

I have the same question too. I think i need to modify the code, but i don't what to do.

[benzhi]

Same here, installing as described on pytorch site (conda install pytorch torchvision cudatoolkit=10.1 -c pytorch) yields (during installation):

  cudatoolkit        pkgs/main/win-64::cudatoolkit-10.1.243-h74a9793_0
  pytorch            pytorch/win-64::pytorch-1.5.0-py3.7_cpu_0
  torchvision        pytorch/win-64::torchvision-0.6.0-py37_cpu

Where there's clearly cpu already in the package names, which can be confirmed after installation:

pytorch                   1.5.0               py3.7_cpu_0  [cpuonly]  pytorch

p.s. CUDA 10.1 was completely installed before creating the env and installing pytorch

[kimcdata]

I had the same problem (Win10, CUDA installed prior to making conda env)

The option to install using pip worked for me (inside a miniconda env, python 3.7.7)

From the pytorch website:

pip install torch===1.5.0 torchvision===0.6.0 -f https://download.pytorch.org/whl/torch_stable.html

[yick2232]

I don't know where to raise an issue, but this continues to be a problem on Windows when Pytorch is installed without CUDA support even though it has been recently fixed for MacOS. Can someone help me?

Change code device = torch.device("cpu") to device = torch.device("cuda")

[sathvikn]

Shalom!

I am new to pytorch, and already installed pytorch in my Mac laptop.

trying to run the wsi_bert.py code, and get this error.
"""
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled
"""
Please help.
Thx a lot!

@Gailsunset I was running the same code from wsi_bert.py, and although conda install -c pytorch torchvision cudatoolkit=10.1 pytorch from the reply right after didn't end up working out, just conda install cudatoolkit led me to get past that issue.

[tejasvi]

In my case, I had cpuonly package installed which needs to be removed.

[bnymnsntrk]

In my case, I had cpuonly package installed which needs to be removed.

how did you delete it?

[amgaweda]

I had the same problem (Win10, CUDA installed prior to making conda env)

The option to install using pip worked for me (inside a miniconda env, python 3.7.7)

From the pytorch website:

pip install torch===1.5.0 torchvision===0.6.0 -f https://download.pytorch.org/whl/torch_stable.html

Just confirming this approach worked for me; Windows 10, using Anaconda's CMD Terminal; CUDA v10

[zm-m]

I don't know where to raise an issue, but this continues to be a problem on Windows when Pytorch is installed without CUDA support even though it has been recently fixed for MacOS. Can someone help me?

I'm having the same problem on an Ubuntu VM. No CUDA installed (because no GPU), and I am 100% sure it is the cpu-only version. I know because I have uninstalled and reinstalled many times (each time making sure all previous files are removed), trying pytorch 1.4, 1.5, and 1.6, all with corresponding torchvision versions. I've also tried editing the code to specify the whole torch.device="cpu" thing but the issue persists. I currently have torch=1.5 with cpu only and torchvision 0.6 installed and would be really grateful to anyone with suggestions.

[OGmenas]

Try this.
conda install -c pytorch torchvision cudatoolkit=10.1 pytorch
Depending on what cuda version you have.

i have a rtx 2060 and it worked fine with this toolkit.

[Malithi-gif]

@Testin1234 are you resolve this one?

[Mwni]

If you've ever ran pip install torch without the -f https://download.pytorch.org/whl/torch_stable.html argument, chances are, you have the corrupt cpu only version in your pip cache. whatever reinstall you do, doesn't matter because pip will just pull the same bad version from cache. to solve:

pip uninstall torch
pip cache purge
pip install torch -f https://download.pytorch.org/whl/torch_stable.html

[BotMichael]

I had the same problem (Win10, CUDA installed prior to making conda env)

The option to install using pip worked for me (inside a miniconda env, python 3.7.7)

From the pytorch website:

pip install torch===1.5.0 torchvision===0.6.0 -f https://download.pytorch.org/whl/torch_stable.html

Thank you so much!!!!!!!!

[lingxuan0520]

I encountered the same problem, I think you should uninstall the cpu version of torch, and change it to the gpu version to solve it

[SharonPeled]

I had the same issue on Linux. I tried almost all the solutions proposed here, but none worked. Finally, I could solve my problem by first uninstalling Torch with pip uninstall torch, and then reinstalling it with Wheel:

pip install torch==1.12.1+cu102 torchvision==0.13.1+cu102 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu102

Then, I got the following, which confirmed that Torch was successfully compiled with CUDA enabled:

>>> import torch
>>> torch.__version__
'1.12.1+cu102'
>>> torch.zeros(1).cuda()
tensor([0.], device='cuda:0')
>>> torch.cuda.is_available()
True

Many thanks for the solution, that's the only solution that worked on my linux machine as well.
Although I'm getting RuntimeError: CUDA error: invalid device function when using nn.parallel.DistributedDataParallel, does anyone else had it? any ideas?

[fengyx8]

这是来自QQ邮箱的假期自动回复邮件。感谢您的来信，邮件我已收到，谢谢！祝您生活愉快！

[meetrais]

If you've ever ran pip install torch without the -f https://download.pytorch.org/whl/torch_stable.html argument, chances are, you have the corrupt cpu only version in your pip cache. whatever reinstall you do, doesn't matter because pip will just pull the same bad version from cache. to solve:

pip uninstall torch
pip cache purge
pip install torch -f https://download.pytorch.org/whl/torch_stable.html

This worked like a charm, Thank you !!!

[geeker-smallwhite]

Hi, I have M2 max chip MacOS with 12 CPU, 38 GPU. I am getting this error :“Torch not compiled with CUDA enabled”. May I ask your help please? Thanks

I heve the same issue in same pc

[fengyx8]

这是来自QQ邮箱的假期自动回复邮件。感谢您的来信，邮件我已收到，谢谢！祝您生活愉快！

[geeker-smallwhite]

Hi, I have M2 max chip MacOS with 12 CPU, 38 GPU. I am getting this error :“Torch not compiled with CUDA enabled”. May I ask your help please? Thanks

I heve the same issue in same pc

you can set runtime mode to resolve this issue

[peterhchen]

#30664 1. Go to NIVIDIA and install cuda toolkit:

nvcc —version

2. Install pytorch

…

On Fri, Sep 29, 2023 at 12:28 AM geeker-smallwhite ***@***.***> wrote: Hi, I have M2 max chip MacOS with 12 CPU, 38 GPU. I am getting this error :“Torch not compiled with CUDA enabled”. May I ask your help please? Thanks I heve the same issue in same pc you can set runtime mode to resolve this issue — Reply to this email directly, view it on GitHub <#30664 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADBSBD47WYWGVZA7QQMXKLLX4Z2BZANCNFSM4JUZTN3A> . You are receiving this because you commented.Message ID: ***@***.***>

[maraujo]

You need to make sure that your cuda driver (downloaded at https://developer.nvidia.com/cuda-downloads) matches with the pytorch-cuda library version that you install with python.

For example, the current snippet at the official pytorch site is:
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia

So, you need to have the 11.8 version of cuda downloaded from nvidia official website and installed.

[fengyx8]

这是来自QQ邮箱的假期自动回复邮件。感谢您的来信，邮件我已收到，谢谢！祝您生活愉快！

[StantheBrain]

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

[liwenyi666]

如果您曾经在pip install torch没有-f https://download.pytorch.org/whl/torch_stable.html参数的情况下运行过，那么您的 pip 缓存中很可能有损坏的仅 cpu 版本。无论你做什么重新安装，都没关系，因为 pip 只会从缓存中提取相同的错误版本。解决：

pip uninstall torch
pip cache purge
pip install torch -f https://download.pytorch.org/whl/torch_stable.html

thank you！！！！！

[fengyx8]

这是来自QQ邮箱的假期自动回复邮件。感谢您的来信，邮件我已收到，谢谢！祝您生活愉快！

[ATkouti]

pip uninstall torch pip cache purge pip install torch -f https://download.pytorch.org/whl/torch_stable.html

I had the same problem and the issue is fixed with the above solution. Just run each command separately and it should solve your problem if it is the same as my problem. 👍 😃 Thank you @Mwni

thank you so much it works for me :)

[fengyx8]

这是来自QQ邮箱的假期自动回复邮件。感谢您的来信，邮件我已收到，谢谢！祝您生活愉快！

[roopeshsn]

I am getting the same error. My laptop doesn't have a dedicated Nvidia GPU. How can I resolve this? Is there any solution?

[fengyx8]

这是来自QQ邮箱的假期自动回复邮件。感谢您的来信，邮件我已收到，谢谢！祝您生活愉快！

[alpaca-farm]

If you've ever ran pip install torch without the -f https://download.pytorch.org/whl/torch_stable.html argument, chances are, you have the corrupt cpu only version in your pip cache. whatever reinstall you do, doesn't matter because pip will just pull the same bad version from cache. to solve:

pip uninstall torch
pip cache purge
pip install torch -f https://download.pytorch.org/whl/torch_stable.html

This one worked for me too! Thanks

[covalspace]

If you've ever ran pip install torch without the -f https://download.pytorch.org/whl/torch_stable.html argument, chances are, you have the corrupt cpu only version in your pip cache. whatever reinstall you do, doesn't matter because pip will just pull the same bad version from cache. to solve:

pip uninstall torch
pip cache purge
pip install torch -f https://download.pytorch.org/whl/torch_stable.html

this didn't work on my mac and now my torch is broken saying module not found but already satisfied so thanks a lot for this

[fengyx8]

这是来自QQ邮箱的假期自动回复邮件。感谢您的来信，邮件我已收到，谢谢！祝您生活愉快！

[1458763783]

How did you install pytorch? It sounds like you installed pytorch without CUDA support. https://pytorch.org/ has instructions for how to install pytorch with cuda support.

I also encountered the same error, this method is worked..

[fengyx8]

这是来自QQ邮箱的假期自动回复邮件。感谢您的来信，邮件我已收到，谢谢！祝您生活愉快！

[StinkyKittyKiss]

If you've ever ran pip install torch without the -f https://download.pytorch.org/whl/torch_stable.html argument, chances are, you have the corrupt cpu only version in your pip cache. whatever reinstall you do, doesn't matter because pip will just pull the same bad version from cache. to solve:

pip uninstall torch
pip cache purge
pip install torch -f https://download.pytorch.org/whl/torch_stable.html

This fixed my problem when encountering this issue while using DeepLabCut (DLC).

[gtimk6]

I use in conda env., reinstall it by commands and solved it.
e.g: conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
pytorch-cuda depends you version.

[Abbsalehi]

device = torch.device("cuda")

This happens once you install pytorch using conda. Instead, install it using:
pip install torch