Incorrect PyTorch 2 + CUDA version

[Dmitry Balabka]

Hello,

M109 image (projects/ml-images/global/images/c2-deeplearning-pytorch-2-0-gpu-v20230615-debian-11-py310) contains the following PyTorch version:
```

$ python -c "import torch; print(torch.__version__)"
2.0.1+cu117

```

While CUDA 11.8 is installed:
```

$ nvcc --versionnvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:33:58_PDT_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0

```

I'm expecting that PyTorch version should support CUDA 11.8.

Could you please provide a correct solution in this case? (installing the correct version of PyTorch or CUDA)

I really appreciate any help you can provide.

[Yuxuan Chen]

Hi Dmitry,

We are aware of this issue and expect to send out a fix next week. Thanks!

--
You received this message because you are subscribed to the Google Groups "google-dl-platform" group.
To unsubscribe from this group and stop receiving emails from it, send an email to google-dl-platf...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/google-dl-platform/55bf8280-7a8e-48b1-ba4a-3cd75ed633e4n%40googlegroups.com.

--

Yuxuan Chen

Deep Learning Environments

[Dmitry Balabka]

Thanks for the quick reply. As a workaround, I've removed the preinstalled torch and installed the correct version from pypi into conda venv:

(base) user@jupyter:~$ pip uninstall  torch torchvision torchaudio
(base) user@jupyter:~$ pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118