NCA Containerization & NGC Catalog 4

Question 1

Which NGC resource type provides pre-trained model weights that can be downloaded and used directly for inference or fine-tuning?

Accepted Answer

NGC Models

Answer

NGC Models are pre-trained model checkpoints available in the NGC Catalog that can be downloaded for direct inference or as starting points for fine-tuning.

Question 2

What is the primary benefit of using multi-stage Docker builds when creating NGC-based custom containers?

Accepted Answer

It reduces final image size by separating build dependencies from runtime artifacts

Answer

Multi-stage Docker builds allow build tools and intermediate files to be excluded from the final image, significantly reducing its size and attack surface.

Question 3

When using NVIDIA Multi-Instance GPU (MIG) with containers, which environment variable specifies the MIG device instance to expose?

Accepted Answer

NVIDIA_VISIBLE_DEVICES

Answer

NVIDIA_VISIBLE_DEVICES is used to specify which GPU or MIG device instance (e.g., MIG-GPU-<uuid>/0/0) is exposed to the container.

Question 4

What type of NGC Catalog resource are 'NVIDIA NIM' microservices classified as?

Accepted Answer

Container Images

Answer

NVIDIA NIM (NVIDIA Inference Microservices) are packaged as optimized container images in the NGC Catalog, making AI model inference deployable via standard container workflows.

Question 5

Which NGC CLI subcommand is used to download a model artifact from the NGC Catalog to local storage?

Accepted Answer

ngc registry model download-version

Answer

The `ngc registry model download-version /:` command downloads a specific model version from the NGC Catalog.

(NCA) NVIDIA-Certified Associate Practice Test

(NCA) NVIDIA-Certified Associate Practice Test

NCA Containerization & NGC Catalog 4