AI Frameworks

What Ships Pre-Installed

RLC Pro AI takes a deliberately focused approach to its framework stack: it ships PyTorch and TorchVision, pre-installed as CIQ-built RPM packages and validated against the kernel, driver, and CUDA versions on the system.

Because the frameworks are system RPMs rather than pip installs, they are cryptographically signed, tracked by the package manager, and updated together with the rest of the OS through dnf update.

Verifying the Stack

PyTorch works from the first boot, with no virtual environment or dependency resolution required:

python3 -c "import torch; print(torch.__version__)"

2.8.0.post12

The pre-installed PyTorch is a CUDA-enabled build:

python3 -c "import torch; print('CUDA build:', torch.version.cuda); print('GPU available:', torch.cuda.is_available())"

CUDA build: 12.8
GPU available: False

The output above is from a host without a GPU: the CUDA 12.8 build is present either way, and torch.cuda.is_available() reports whether a supported NVIDIA GPU is visible to the driver. On GPU hardware it returns True; without one, PyTorch transparently runs on the CPU. No separate CPU-only package is needed.

Confirm the packages are managed by RPM:

rpm -q python3-torch python3-torchvision

python3-torch-2.8.0-12.el9_ciqai.x86_64
python3-torchvision-0.23.0-1.el9_ciqai.x86_64

The el9_ciqai tag identifies packages built by CIQ for the RLC Pro AI product.

Additional Packages Available via DNF

The rlc-9-ai repository carries supporting libraries beyond the pre-installed set, all built and signed by CIQ. Notable examples:

sudo dnf install python3-onnx

Browse the full set with:

sudo dnf repoquery --disablerepo="*" --enablerepo="rlc-9-ai.x86_64"

Installing Additional Frameworks with pip

For frameworks and libraries not packaged by CIQ, use pip inside a virtual environment. The base image intentionally ships without pip, so install it first:

sudo dnf install python3-pip

Then create a virtual environment for your project. Using --system-site-packages gives the environment access to the pre-installed, validated PyTorch build instead of downloading a second copy:

python3 -m venv --system-site-packages ~/venvs/myproject
source ~/venvs/myproject/bin/activate
pip install transformers

Updating the Stack

Framework updates arrive through the same channel as every other package, tested against the kernel and driver versions they ship with:

sudo dnf update

There is no separate framework update procedure, and no risk of pip upgrading PyTorch out from under a validated driver stack.