As you may know, the torch package is a modified version and already installed in the NGC PyTorch container. And the version number is really wired, for example, 2.5.0a0+e000cf0ad9.nv24.10. Considering that we may use different versions of the containers, I want to know if there is any best practice to manage the dependencies with pdm.
I tried to use the following command to generate a lock file, but it complains with the following error:
$ pdm lock --strategy no_cross_platform --global --project . --prod --update-reuse-installed
⠇ 0:00:10 Resolve for environment (>=3.10) None
See /home/huxuan/.local/state/pdm/log/pdm-lock-bj6ljqpb.log for detailed debug log.
[CandidateNotFound]: No candidate is found for `torch` that matches the environment or hashes
WARNING: Add '-v' to see the detailed traceback
I also tried to exclude torch package as mentioned in #3578, but it does not seem to work even on my laptop.
If there is any further information I should provide, just let me know.
As you may know, the
torchpackage is a modified version and already installed in the NGC PyTorch container. And the version number is really wired, for example,2.5.0a0+e000cf0ad9.nv24.10. Considering that we may use different versions of the containers, I want to know if there is any best practice to manage the dependencies with pdm.I tried to use the following command to generate a lock file, but it complains with the following error:
I also tried to exclude
torchpackage as mentioned in #3578, but it does not seem to work even on my laptop.If there is any further information I should provide, just let me know.