Skip to main content

Environments

LLM Launcher → Environments gives you popular deep learning and machine learning frameworks as ready-to-run containers, with no installation on your side. Both NVIDIA NGC containers and Cordatus containers are available, optimised for the hardware you have registered.

An environment is a development image — PyTorch, TensorFlow, CUDA, TensorRT and so on — rather than an inference engine. It uses the same launch flow as an application, minus the model-selection step.

Browsing the catalogue, the launch wizard end to end, and the notebook URL after start

 

The Environments page​

Two sections:

  • Most Popular — a slider of the most-used environments, with ‹ › navigation and slide indicators. The count is shown beside the heading.
  • All Environments — the rest, as a grid, with View All opening the complete list.

Each card shows the environment's logo, its name, a short description and up to three labels (the rest behind +{n}). Clicking a label filters by it; clicking the card, or List Environments, opens its versions.

The refresh icon beside the page title reloads the catalogue.

Supported environments​

Cordatus currently offers, among others:

  • Python
  • OpenCV
  • PyTorch
  • TensorFlow
  • ONNX
  • TensorRT
  • Torchvision
  • Caffe
  • Chainer
  • CUDA
  • cuDNN
  • DALI
  • DLib
  • MS CNTK
  • MXNet
warning

Training models on the NVIDIA Jetson platform may lead to overheating and can damage the Jetson module. Use NVIDIA RTX workstations for model training.

Choosing a version​

List Environments opens the version list for that environment. The page gives you:

  • search;
  • Filter by device type — Server - Workstation, Jetson, DGX Spark;
  • for Jetson, an additional Select Jetson group… filter;
  • removable filter chips and Clear All.

Some environments publish different versions for different hardware — for example L4T builds for NVIDIA Jetson devices. Only what your device can run is offered.

View Details opens the environment, with the same four tabs as an application: README/Description, Versions, Containers and (for engines) Models. See the Launch Guide for what each tab holds.

Starting an environment​

Start Single Device Environment opens the launch wizard:

  1. Select Device — a connected device.
  2. Select Version — a compatible image, marked Downloaded or Will Download.
  3. Advanced Configuration — the rail.

Because a development environment has no model, the rail is shorter than for an engine. What you configure is:

Environment Name​

A custom name for the container. Left blank, Cordatus generates one.

Compute​

  • GPUs — pick individual GPUs or switch on Use all. At least one is required. On a multi-GPU device this is how you isolate work per environment.
  • Host limits — Auto keeps the Host Reserved CPU and RAM back for the system; Custom lets you set the caps yourself.
tip

Keep the Host Reserved values Cordatus sets automatically. They are what keeps the device responsive while the container is busy.

Overrides​

The same tabs as an application, but only the ones this image has something for — a plain environment usually shows just Container and Notebook:

  • Container — docker run options: --volume, --network, --env, --restart, --device and the rest, through a form rather than a command line.
    • Ports — Cordatus suggests a free port and warns when one is already in use.
    • Volumes — a built-in file browser over the device's real disks, where you can navigate, create folders and bind them into the container.
  • Environment — environment variables, including values taken from your token vault.
  • Engine flags — where the image publishes any.
  • Notebook — Jupyter.

Rows the image fixes are shown with a lock and cannot be removed.

Notebook (Jupyter)​

Jupyter runs inside the same container, so notebooks share the mounted folders and the GPUs.

How notebooks are protected offers three modes:

ModeWhat it means
PasswordA Jupyter password you type. Asked once per browser when opening the notebook.
TokenAn Access token generated for you, with regenerate and copy buttons. It appears again in the environment's Ports tab after start.
No authNo authentication at all. The panel turns to Not protected.
caution

An unprotected notebook gives anyone who can reach the host on that port a shell-capable notebook. Use it only on an isolated network.

Review & start​

The last pane shows the summary, the exact command that will run, and every container that will be created — then Start Environment.

After it starts​

  1. If the image is not on the device yet, Cordatus asks first — "Environment is not exist. Would you like to download the environment to {device}?" — and Download Now pulls it; the container shows as Downloading.
  2. With a notebook enabled, the Jupyter Token dialog shows the token once and copies it to your clipboard. Save it before closing: it is not shown again.
  3. The container appears on the Containers page.
  4. Open the details panel → Ports for the local address, the notebook URL and any public URL.

Starting from your own image​

An environment can also be started from the Private Registry, from a tag you pushed yourself. That path skips the version step — the tag is the version.