Closed bad-meets-joke closed 1 year ago
TLDR: Sorry, I dunno. Please develop on up-to-date enviornments(torch 112, vision 0.13).
Longer story Ugh, the vision part on the ablative is not installed by recent, so I don't have the installation packages either.
And the guide is intended to set up an up-to-date environment but not the exact same one with my ablative machine. The machine has more than one users so I can't simply update it on my own...
My suggestion is to update everything, specifically using pytorch 112 and vision 0.13.
FYI, the vision on the ablative machine has version string '0.8.2+cu101'
Still, I highly recommend to develop on up-to-date environment, which is more future proof.
TLDR: Sorry, I dunno. Please develop on up-to-date enviornments(torch 112, vision 0.13).
Longer story Ugh, the vision part on the ablative is not installed by recent, so I don't have the installation packages either.
And the guide is intended to set up an up-to-date environment but not the exact same one with my ablative machine. The machine has more than one users so I can't simply update it on my own...
My suggestion is to update everything, specifically using pytorch 112 and vision 0.13.
Ok,I will change to pytorch1.12.0, so how to get the vision-0.13.tar.gz as decribed in the manual.pdf?
you can get it from the official site :-) They seem to have fixed the ffmpeg bug.
https://github.com/pytorch/vision/releases
Alternatively you can find it in https://github.com/lancercat/make_env
they should be the same.
Thanks, I run on the Pytorch1.7.1+cu10.1 due to the limitation of Nvidia Driver version, and directly use conda install torchvision==0.8.2
, it finally works.
Thanks for your feedback! I pinned this issue as I think pretty many people using cent or ubuntu may find this helpful.
Could you share the link to download the torchvision code, i.e. vision-xxx.tar.gz described in manul.pdf, of which the version is also compatible with pytorch1.7.1+cu10.1?