NVIDIA NGC API Key

Why do you need to configure an NVIDIA NGC API Key to deploy NVIDIA NIM?

You need to configure an **NVIDIA NGC API Key** to deploy **NVIDIA NIM (NVIDIA Inference Microservices)** because the key: ### ✅ Grants access to NGC-hosted assets: NIMs are containerized models hosted on the **NVIDIA GPU Cloud (NGC)** platform. The API key does the following: 1. **Authenticates your identity** It proves to NGC that you’re a verified user and have permissions to access NVIDIA's enterprise or community model containers. 2. **Authorizes access to private or licensed content** Some models and NIMs require licenses or special access (e.g., enterprise models like NeMo, BioNeMo, or Clara). Your key is tied to your entitlements. 3. **Enables pull of containers and model weights** When deploying NIM via Docker or Kubernetes, the system pulls the required containers and model checkpoints from `nvcr.io`. This access requires a valid NGC API key. --- ### 🔐 Without the API key: - You cannot pull the container images from `nvcr.io`. - The deployment will fail because it can't access the required model or runtime environments. --- ### 📍How to configure: 1. Create an account on [NGC](https://ngc.nvidia.com). 2. Go to **Setup > API Key** in your NGC dashboard. 3. Generate and download the key. 4. Add it as an environment variable or to your `~/.ngc/config` file depending on your deployment method. ```bash export NGC_API_KEY=your_key_here ``` --- ### 🔁 Summary: You need an NGC API key because **NIM relies on proprietary NVIDIA model containers stored on NGC**, and the key is the gatekeeper that controls and secures your access.

Before installing the NVIDIA NIM with Helm, what Kubernetes resources should you create to deploy the microservice on? (choose all that apply). Choose as many answers as you see fit. A new GKE cluster GPU node pool Namespace A deployment configuration

To deploy **NVIDIA NIM** with Helm, you should **prepare the following Kubernetes resources**: --- ### ✅ **Correct answers:** 1. **✅ A new GKE cluster** - **Required** if you don’t already have a Kubernetes cluster. - GKE (Google Kubernetes Engine) is just one example—any K8s cluster with GPU support works. 2. **✅ GPU node pool** - **Essential**: NIM models require **GPU acceleration**. - Your cluster must include nodes with NVIDIA GPUs and the correct drivers (usually via the NVIDIA Device Plugin). 3. **✅ Namespace** - Not strictly required, but **strongly recommended** to isolate the NIM deployment. - Helm charts often expect or allow configuration of a namespace. 4. **✅ A deployment configuration** - Helm itself creates Kubernetes deployments, but you typically supply **custom configuration values** (e.g., `values.yaml`) to define: - Model type - Resources (GPU/CPU/memory) - Storage - Networking (ports, ingress, etc.) --- ### ❌ Trick note: Even though **Helm automates** much of the deployment, you **must** set up the underlying **infrastructure and resources** like the GPU pool and optionally namespace/configs beforehand. --- ### ✅ Final Answer: - ✅ A new GKE cluster - ✅ GPU node pool - ✅ Namespace - ✅ A deployment configuration

To pull the NIM container image from the nvcr.io registry, what type of resource should you create in your namespace using an NGC API key?

To pull the NIM container image from the `nvcr.io` registry using your **NGC API key**, you should create a: --- ### ✅ **Kubernetes Secret of type `docker-registry`** This secret stores your **NGC API key as Docker credentials**, allowing Kubernetes to authenticate with NVIDIA’s private container registry (`nvcr.io`) when pulling the image. --- ### 🔧 Command to create it: ```bash kubectl create secret docker-registry ngc-secret \ --docker-username='$oauthtoken' \ --docker-password=<YOUR_NGC_API_KEY> \ --docker-server=nvcr.io \ --namespace=<YOUR_NAMESPACE> ``` > 🔒 `'$oauthtoken'` is a special username required by NGC. > 🔐 `<YOUR_NGC_API_KEY>` is the actual API key string you downloaded from NGC. --- ### 📌 After creating the secret: You reference it in your **Helm values file** or Kubernetes deployment under `imagePullSecrets`, like so: ```yaml imagePullSecrets: - name: ngc-secret ``` --- ### ✅ Final Answer: **A Kubernetes image pull secret (type `docker-registry`)** containing your NGC API key.

In the nim_custom_value.yaml file, what is the purpose of the image.repository key? To set the NIM deployment version. To define which model container we want to deploy. To name the Kubernetes secret for pulling images. To define the Helm chart location. Which command is used to install the NIM using a Helm chart, a custom configuration file, and a specific namespace? gcloud container clusters create kubectl apply helm install docker run How do you make the NIM service accessible to your local Cloud Shell terminal for testing? Use kubectl port-forward. Assign a public IP to the pod. Edit the cluster's firewall rules. SSH into the GKE node.

Here's a breakdown of **all three questions** with the correct answers and explanations: --- ### **Q1: In the `nim_custom_value.yaml` file, what is the purpose of the `image.repository` key?** ✅ **Correct Answer:** > **To define which model container we want to deploy.** 🧠 **Explanation:** - The `image.repository` key in `nim_custom_value.yaml` specifies the **container image path** (e.g., `nvcr.io/nvidia/nim-gpt`). - This tells Helm **which NIM model** you are deploying (e.g., LLMs, ASR models, etc.). --- ### **Q2: Which command is used to install the NIM using a Helm chart, a custom configuration file, and a specific namespace?** ✅ **Correct Answer:** > **`helm install`** 🧠 **Explanation:** - `helm install` is used to deploy applications using a Helm chart. - You include: - The Helm chart name - The `-f nim_custom_value.yaml` flag for your custom config - The `--namespace` flag for targeting a specific namespace **Example:** ```bash helm install nim nvidia/nim \ -f nim_custom_value.yaml \ --namespace nim-namespace ``` --- ### **Q3: How do you make the NIM service accessible to your local Cloud Shell terminal for testing?** ✅ **Correct Answer:** > **Use `kubectl port-forward`.** 🧠 **Explanation:** - `kubectl port-forward` tunnels a local port on your Cloud Shell (or dev machine) to a port on the Kubernetes service or pod. - This is the **quickest and safest** way to test locally without modifying firewall rules or exposing public IPs. **Example:** ```bash kubectl port-forward service/nim-service-name 8000:80 \ --namespace nim-namespace ``` --- ### ✅ Summary of Correct Answers: 1. **image.repository** → ✅ *To define which model container we want to deploy* 2. **Install NIM** → ✅ *`helm install`* 3. **Access from Cloud Shell** → ✅ *Use `kubectl port-forward`*