Exercise guide — refer to the official documentation for full details.
Environment assets define the container image and runtime configuration for your workloads.
Navigate to Workload Manager > Assets > Environments
Click + New Environment
Fill in the following:
| Field | Value |
|---|---|
| Scope | alpha-project-1-gpu |
| Name | mnist-jupyter-lab-dev |
| Image URL | nvcr.io/nvidian/demo-pytorch-jp-example:25.01-py3 |
Select Standard and Workspace as the workload architecture and type
Click on Tools
Click on +Tool
Select Jupyter
Click on Runtime Settings
Click on + Command and Arguments
Fill in:
| Field | Value |
|---|---|
| Command | jupyter-lab |
| Arguments | --NotebookApp.base_url=/${RUNAI_PROJECT}/${RUNAI_JOB_NAME} --NotebookApp.token='' --ServerApp.allow_remote_access=true --allow-root --port=8888 --no-browser |
Click on Create Environment
Navigate to Workload Manager > Assets > Environments
Click + New Environment
Fill in the following:
| Field | Value |
|---|---|
| Scope | alpha-project-1-gpu |
| Name | mnist-standard-training |
| Image URL | nvcr.io/nvidian/demo-pytorch-standard-example:26.01-py3 |
Select Standard and Training as the workload architecture and type
Click on Runtime Settings
Click on + Command and Arguments
Fill in:
| Field | Value |
|---|---|
| Command | ./run.sh |
Click on Create Environment
Navigate to Workload Manager > Assets > Environments
Click + New Environment
Fill in the following:
| Field | Value |
|---|---|
| Scope | omega-project-4-gpus |
| Name | pytorch-distributed-training-example |
| Image URL | nvcr.io/nvidian/demo-pytorch-ddp-example:24.07-py3 |
Select Distributed as workload architecture
Select PyTorch as the framework for the distributed workload
Select Training as the workload type
Click on Runtime Settings
Click on + Command and Arguments
Fill in:
| Field | Value |
|---|---|
| Command | ./run.sh |
Click on Create Environment
Check the Workload Manager > Assets > Environments page — all environments should be listed and available for workload creation.
!!! tip
Pin image tags to specific versions (e.g., 2.0.1) rather than using latest for reproducibility.