search

iot-salzburg / gpu-jupyter

GPU-Jupyter: Leverage the flexibility of Jupyterlab through the power of your NVIDIA GPU to run your code from Tensorflow and Pytorch in collaborative notebooks on the GPU.
Apache License 2.0
698 stars 231 forks source link

PyTorch 2 needs CUDA 11.7+ #119

Closed hlapp closed 9 months ago

hlapp commented 1 year ago

The latest CUDA version available among the prebuilt images on Docker Hub seems to be 11.6. However, apparently PyTorch2, which seems to be the version of PyTorch installed into the container, needs 11.7+. Are there plans to add pre-builds for later CUDA versions?

hlapp commented 1 year ago

Just to be clear, thinking that PyTorch2 is installed in the container was a mistake of mine.

mathematicalmichael commented 11 months ago

hm. i've been using torch 2 since release and don't recall rebuilding my images. are you able to just pip install it?

hlapp commented 11 months ago

The installation is ultimately unsuccessful because of the CUDA dependency. Of course, this only matters if using on GPU.

benz0li commented 9 months ago

@hlapp You may be interested in b-data's/my GPU accelerated JupyterLab docker stacks.

(currently) Based on nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04; including code-server – aka VS Code in the browser.

ChristophSchranz commented 9 months ago

Hi, new tags are available such as v1.5_cuda-12.0_ubuntu-22.04_python-only and v1.5_cuda-11.8_ubuntu-22.04_python-only which are based on 12.0.1-cudnn8-devel-ubuntu22.04 rsp. 11.8.0-cudnn8-devel-ubuntu22.04. Both are built with with PyTorch 2 and the latest compatible Tensorflow version.

Thiss issue should be solved with PR https://github.com/iot-salzburg/gpu-jupyter/pull/124