For my project, I need to use pytorch, torchtext, and torchdata. However, I get the following error with the following versions
torch==1.12.0+cu102
torchtext==0.13.0
torchdata==0.4.0
NVIDIA A100-PCIE-40GB with CUDA capability sm_80 is not compatible with the current PyTorch installation
My system is a remote machine with Fedora Linux that I can only access via SSH. I do not have root access or full priviliges, so no sudo shananigans please. Unfortunately, I cannot try the following option: How does one use Pytorch (+ cuda) with an A100 GPU? because I cannot find the corresponding torchdata and torchtext versions (they only seem to start at torch 1.11.0). How do I solve this issue? Many thanks for your help.
I also have some issues installing pytorch with a specific cuda version. For example:
pip install torch==1.11.0+cu113
Defaulting to user installation because normal site-packages is not writeable
ERROR: Could not find a version that satisfies the requirement torch==1.11.0+cu113 (from versions: 1.11.0, 1.12.0, 1.12.1, 1.13.0, 1.13.1, 2.0.0, 2.0.1)
ERROR: No matching distribution found for torch==1.11.0+cu113
An alternative solution would be to get rid off torchtext altogether. Currently, I need it to load pretrained embedding vectors:
glove = torchtext.vocab.GloVe(name='twitter.27B', dim=100)
Loading the file "glove.twitter.27B.100d.txt" would obviate my need for torchtext, so I will accept that as an alternative solution.
I managed to solve it by installing a newer cuda version (> 1.10)
pip install torch==1.12.0+cu116 torchvision==0.13.0+cu116 torchaudio==0.12.0 --extra-index-url https://download.pytorch.org/whl/cu116 --no-cache-dir
--no-cache-dir was needed because the process was killed every time I tried without.