Its very unclear at the moment how Rye can be used with existing docker containers for example shipped by nvidia. Many engineers start with a container and install packages that respect the nvidia optimized / accelerated package versions therein.
FROM nvcr.io/nvidia/pytorch:23.10-py3 as builder
At some point the rye toolchain can be set to the existing toolchain
RUN rye toolchain register --name nvidia /usr/bin/python
Its unclear though what should happen after the pinning and syncing past registration to get to the desired behavior:
When you do rye add torch && rye sync, should use the existingt torch installation rather than a new independent installation into a .venv.
There is a way to use pip-tools and pip constraints to achieve the above but I am wondering if there are others that faced this issue and managed to create a workflow based on Rye.
Its very unclear at the moment how Rye can be used with existing docker containers for example shipped by nvidia. Many engineers start with a container and install packages that respect the nvidia optimized / accelerated package versions therein.
FROM nvcr.io/nvidia/pytorch:23.10-py3 as builder
At some point the rye toolchain can be set to the existing toolchain
RUN rye toolchain register --name nvidia /usr/bin/python
Its unclear though what should happen after the pinning and syncing past registration to get to the desired behavior:
When you do
rye add torch && rye sync
, should use the existingt torch installation rather than a new independent installation into a .venv.There is a way to use pip-tools and pip constraints to achieve the above but I am wondering if there are others that faced this issue and managed to create a workflow based on Rye.