Cuda too old
2023-12-19T14:21:37.212836490Z The NVIDIA driver on your system is too old (found version 11070). Please update your GPU driver by downloading and installing a new version from the URL: http://www.nvidia.com/Download/index.aspx Alternatively, go to: https://pytorch.org to install a PyTorch version that has been compiled with your version of the CUDA driver.: str
2023-12-19T14:21:37.214841659Z Traceback (most recent call last):
2023-12-19T14:21:37.214853339Z File "/stable-diffusion-webui/modules/errors.py", line 84, in run
2023-12-19T14:21:37.214858479Z code()
2023-12-19T14:21:37.214861749Z File "/stable-diffusion-webui/modules/devices.py", line 63, in enable_tf32
2023-12-19T14:21:37.214865109Z if any(torch.cuda.get_device_capability(devid) == (7, 5) for devid in range(0, torch.cuda.device_count())):
2023-12-19T14:21:37.214868209Z File "/stable-diffusion-webui/modules/devices.py", line 63, in <genexpr>
2023-12-19T14:21:37.214871259Z if any(torch.cuda.get_device_capability(devid) == (7, 5) for devid in range(0, torch.cuda.device_count())):
2023-12-19T14:21:37.214874339Z File "/opt/conda/lib/python3.10/site-packages/torch/cuda/init.py", line 435, in get_device_capability
2023-12-19T14:21:37.214877390Z prop = get_device_properties(device)
2023-12-19T14:21:37.214880450Z File "/opt/conda/lib/python3.10/site-packages/torch/cuda/init.py", line 449, in get_device_properties
2023-12-19T14:21:37.214883470Z _lazy_init() # will define _get_device_properties
2023-12-19T14:21:37.214886600Z File "/opt/conda/lib/python3.10/site-packages/torch/cuda/init.py", line 298, in _lazy_init
2023-12-19T14:21:37.214891770Z torch._C._cuda_init()...
Download the latest official NVIDIA drivers
Download the latest official NVIDIA drivers
PyTorch
PyTorch
8 Replies
is it my dependency problem?
Is this an error that you saw, which template are you using?
yeah
im using my custom template
FROM pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime
RUN pip install torchvision torchaudio --prefer-binary
RUN pip install --no-cache-dir xformers pyngrok --no-dependencies --prefer-binary
i guess these are the lines that you need, on some machines my template works
like 90%ish of the time it works
sometimes it gets this error
When you launch the pod are you using the CUDA filter to select the minimal version?
im using serverless
Got it, let me check what the status is with adding that feature
oh ya there is something related with the machine's nvidia driver
i only get the problem in A4500 A4000 i think