AssertionError: Torch not compiled with CUDA enabled

[ngrivv]

Getting this error when trying to run the script

[martintomov]

1. Check if your system has CUDA installed:

   nvcc --version
   

2. Run the following .py code to verify if PyTorch can detect your GPU:

   import torch
   print(torch.cuda.is_available())
   

   If it returns False, PyTorch isn't detecting CUDA, meaning that the installed version might be CPU-only.

3. Install the Correct Version of PyTorch:

   pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
   
   

4. Verify Installation:

   import torch
   print(torch.cuda.is_available())
   print(torch.cuda.get_device_name(0))
   

5. Try running the config again

[suifeng1221]

the same mistake，still not solved

[ngrivv]

Hi, turns out the issue was that I was using the wrong start command (on Windows) so use python run.py config/whatever_you_want.yml instead of python3 run.py config/whatever_you_want.yml

[martintomov]

Hi, turns out the issue was that I was using the wrong start command (on Windows) so use python run.py config/whatever_you_want.yml instead of python3 run.py config/whatever_you_want.yml

This exists to distinguish between Python 2.x and Python 3.x installations on systems where both versions are available. Nowadays, Python 2 is hardly ever used, so the best way is to check which one you have is with --version:

python --version
python3 --version
pip --version
pip3 --version

[deeplearner5]

Try pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118 Looks like the wrong torch is being picked up, that version should include CUDA 11.8. Just saw this also mentioned on main page, but with cu121.