-
Notifications
You must be signed in to change notification settings - Fork 95
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Poetry install of torch+cuda build variants not possible through a clearml agent #88
Comments
Hi @hamishc , If you're using the agent in docker mode, you can specify custom commands to execute as a bash script after the docker container has been started and python has been installed. Take a look here: https://github.com/allegroai/clearml-agent/blob/6c5087e425bcc9911c78751e2a6ae3e1c0640180/docs/clearml.conf#L145 |
The workaround above doesn't work if run before |
Hi @hamishc, what would be the command run in such an entrypoint? |
To fix this issue I'd just need to run the pip install on the poetry venv above: More generally, we might want to run anything that needs post-poetry execution - like poe (https://github.com/nat-n/poethepoet) with |
I think this was just introduced in poetry - it is already supported, see python-poetry/poetry#2613 and https://stackoverflow.com/a/70342258 🙂 |
If you use the marker / url constraint, you get this error when you try to install the torchvision build variant:
|
Hi @hamishc, After some digging around, see here - as this is an actual poetry bug/issue (which I think they are working on resolving) - I would hate to add another layer that might break it... |
Poetry doesn't yet support torch+cuda build variants (python-poetry/poetry#2613), and they've indicated that it may take a while to implement (python-poetry/poetry#4221).
The current workaround is to just run
poetry run python -m pip install torch==1.8.0+cu111 torchvision==0.9.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html
after runningpoetry install
, but when running through a clearml agent this isn't possible (as far as I can tell).Is there a workaround for this issue that's usable in clearml, or is it not currently possible to use poetry + torch/cuda build variants + clearml?
The text was updated successfully, but these errors were encountered: