NCP-AI Questions and Answers

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

When deploying Nutanix Enterprise AI on managed Kubernetes services like Google Kubernetes Engine (GKE) or Azure Kubernetes Service (AKS), the management of GPU drivers is typically handled by theNVIDIA GPU Operatoror the cloud provider's pre-configured node images.

Driver Management:The deployment workflow involves installing the NVIDIA GPU Operator, which automatically manages the driver container, container toolkit, and device plugin.

Manual Steps:Consequently, the administrator shouldSkip installing the GPU drivermanually on the worker nodes during the preparation phase. Installing drivers manually can create conflicts with the drivers deployed by the GPU Operator or the underlying managed OS image.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

In the Nutanix Enterprise AI "Models" dashboard, the Pending status has a specific definition related to resource availability.13 The documentation states: "Pending: The system is waiting for the resources required to save the LLM.14 For example, if the required storage space is not available, Nutanix Enterprise AI maintains a Pending status until the storage space becomes available."15

This indicates that the import process has been accepted but cannot proceed to the "Processing" (downloading) phase because the cluster lacks the necessary underlying resources (specifically storage capacity in the PVC/Files support) to ingest the model.

If the token were unauthorized, the status would typically result in aFailedstate.

If the model were custom/unvalidated, it would still process if resources were available.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

The symptoms described (CPU usage at 100%, pods waiting for resources/pending state, critical infrastructure health) indicate that the underlying Kubernetes cluster has run out of compute capacity (nodes).

Nutanix Enterprise AI (NAI) runs as an applicationon top ofa Kubernetes platform, typically theNutanix Kubernetes Platform (NKP). NAI itself manages the scaling ofinference endpoints(pods), but it relies on the underlying platform to provide the physical or virtual compute nodes. To resolve a cluster-wide resource exhaustion where CPU is 100% saturated, the administrator must expand the cluster's capacity. The most appropriate and automated way to handle this is toenable automatic resource scaling (Cluster Autoscaler) within NKP. This allows the platform to automatically provision additional worker nodes when pending pods cannot be scheduled due to resource constraints, thereby satisfying the demand for the Custom Chatbot service.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

According to the endpoint creation workflow in the Nutanix Enterprise AI documentation, an endpoint is essentially a deployment of a specific model. Therefore, the prerequisite for creating an endpoint is having a model available in the local registry.

Model Status:The system explicitly requires that the imported Large Language Model (LLM) be in theActive(or Ready) state before it can be selected in the "Create Endpoint" wizard.

Workflow:When an administrator imports a model (whether from Hugging Face or manually), it transitions through statuses like "Processing" (downloading) or "Pending" (waiting for resources). It is only when the status reachesActivethat the model is fully validated, stored, and available to be served as an endpoint. An "Uploaded" status typically refers to a manual file transfer that has not yet been processed or validated for deployment.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

According to the Nutanix Enterprise AI "User Roles" and privileges documentation, the AI/ML User is the least privileged role that still retains the capability to import Large Language Models (LLMs).

The permissions table explicitly lists "Import an LLM" asYesfor the AI/ML User role. However, a key distinction is that an AI/ML User can only view, manage, and delete the LLMs that they have imported themselves, whereas an AI/ML Admin or Super Admin has broader visibility and management rights over models imported by other users. Since the question asks for theleastprivileged role required to perform the import action,AI/ML Useris the correct answer.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

According to the "Sizing Requirements" section of the Nutanix Enterprise AI deployment guide, specific minimum persistent storage values are assigned to each internal component of the NAI control plane.

nai-db Requirements:The nai-db component, which is responsible for storing application data in a PostgreSQL database, explicitly requires a minimum of4 GiBof persistent storage.

Storage Access Mode:This storage is provisioned using a ReadWriteOnce (RWO) access mode, typically backed by Block storage.

Comparison:To differentiate from other components:

The nai-api (Model) component typically requires 20 GiB.

The prometheus-nai component also typically requires 20 GiB.

The 2 TiB requirement mentioned in other contexts usually refers to the shared storage (NFS/Object) required for holding the actual Large Language Model (LLM) weights, not the internal nai-db database.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

According to the Role-Based Access Control (RBAC) definitions within the Nutanix Enterprise AI (NAI) platform, there is a strict segregation of resources to ensure privacy and security in multi-tenant environments.

User Isolation:While anAI/ML Userhas the privileges to create, manage, and delete theirownendpoints and generate API keys for applications they are developing, they do not have global visibility into the private resources of others.

Model Visibility:An AI/ML User can view and deployPre-validated Models(provided globally by the Administrator) orCustom Modelsthat they have imported themselves (if the "Allow Manual Upload" permission is granted).1However, theycannot viewCustom Models that have been imported by other AI/ML Users. These models are scoped to the user who created them to preventing unauthorized access to proprietary or sensitive model weights uploaded by different teams.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

For a secure deployment of Nutanix Enterprise AI, particularly regarding model storage and container registries, the platform enforces strict security protocols.

Disable HTTP (Option A):NAI requires secure connections. The registry should not expose insecure HTTP endpoints; it must be configured to serve traffic exclusively over HTTPS.

Trusted Certificate (Option C):While self-signed certificates can be technically made to work with extensive workaround configurations (adding CA to trust stores on all nodes), the "deployment prerequisites" and supported security stance for NAI state that the registry must be secured with atrusted TLS/SSL certificate(signed by a valid Certificate Authority). This ensures that the underlying Kubernetes cluster and NAI services can validate the registry's identity without connection errors or security warnings during image pulls.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

To install Nutanix Enterprise AI (NAI), specifically when pulling the required container images and Helm charts from the Nutanix repository, authentication is required.2

Access Token Source:The administrator must log in to theNutanix Portal(specifically the Software Downloads or Licenses section) to generate or retrieve theAccess Token(often referred to as the API Key for the registry).3

Usage:This token is then used as the DOCKER_API_KEY or password in the Helm install command (helm install ... --set imagePullSecret.password=...) to authenticate against the ntnx-charts or specific container registry.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

TheView Sample Request(or "View Sample Code") feature in the Endpoint Details or List view is designed to facilitate immediate connectivity for developers.

Content:Clicking this link opens a dialog that displays a code snippet (formatted for cURL, Python, etc.) containing exactly two critical pieces of information: theEndpoint URLand the associatedAPI Keyrequired to authenticate against that endpoint.

Purpose:This allows administrators or developers to copy a ready-made command that verifies connectivity without having to manually construct the URL or hunt for the API key in a separate menu.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

Nutanix Enterprise AI (NAI) is explicitly architected to support and validate models from two primary ecosystem sources/formats: NVIDIA NIM (NVIDIA Inference Microservices) and Hugging Face.

The NAI "Import Model" interface allows administrators to select models directly from theHugging Face Model Hub(supporting various model architectures hosted there) or from theNVIDIA NGC Catalog(importing NVIDIA NIMs). The documentation repeatedly refers to "Importing a Large Language Model from Hugging Face" and "Importing NVIDIA NIMs" as the core supported workflows. While other formats like ONNX exist, the official platform support forimporting and packagingfocuses on the NIM containers and Hugging Face repositories.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

In Nutanix Enterprise AI, API keys are securely managed and are strictly associated with the user who created them. For security reasons, once an API key is created and displayed, it is masked and cannot be viewed again in the dashboard.

If an AI/ML Admin attempts to test an endpoint and relies on an API key created byanotherAI/ML Admin, they will be unable to retrieve or view that key (as it is masked for all users except the creator at the moment of creation). Furthermore, if the client application is using a key that the current admin cannot validate or manage, access may fail if that key was revoked or not shared securely offline. The visibility and management of API keys are scoped to the creator; therefore, the testing admin should generate their own API key and assign it to the endpoint to ensure successful access.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

This scenario describes a case where the Nutanix Enterprise AI infrastructure is functioning correctly (indicated by the 'Active' state, 'Succeeded' status, and successful sample requests), but thecontentgenerated by the model is unsatisfactory.

Distinguishing Infrastructure vs. Model Capability:Since the infrastructure is healthy (ruling out Option A) and there are no reports of timeouts or slowness (ruling out Option B), the issue lies with thesuitability of the selected Large Language Model (LLM)for the specific business task.

Model Selection Best Practices:The NAI documentation advises that not all models are suitable for all tasks. A model might return "accurate" responses to simple "Hello" prompts (as seen in sample tests) but fail at complex domain-specific queries if it was not trained on relevant data.

Correct Remediation:The administrator should advise the developers to consult theModel Cardor documentation (typically hosted on Hugging Face) to verify the model's training data, parameters, and intended use cases. It is likely the selected model is too generic or small for a specialized customer service role. This step is the necessary precursor to deciding if they need to switch to a more capable model or implement Retrieval-Augmented Generation (RAG) to provide external context.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

Based on the Nutanix Enterprise AI (NAI) role-based access control (RBAC) and model management documentation:

Manual Upload Permissions:The action shown in the exhibit corresponds to aManual Import(or "Manual Upload") of a model using an infrastructure storage source (NFS File Share). This workflow is typically used for "Dark Site" (air-gapped) environments or custom model ingestion.

Role Restrictions:By default or in secure configurations, theAI/ML Userrole is often restricted from performing manual uploads from infrastructure sources like NFS shares or S3 buckets. This capability generally requires theAllow Manual Uploadsetting to be explicitly enabled by an admin, or the action must be performed by anAI/ML Admin.

Infrastructure Access:Connecting to a "File Share" via NFS (as shown with fs01.nutanix.edu) involves mounting infrastructure storage to the NAI cluster. Standard AI/ML Users typically do not have the underlying privileges to define these storage mounts due to security risks. They are primarily consumers of models or importers from public hubs (like Hugging Face) if allowed, rather than infrastructure architects.

Conclusion:The import failed because the user attempting the action (an AI/ML User) lacked the necessary permissions to execute a manual import from a backend file share.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

To onboard a custom or "unvalidated" model directly from Hugging Face into the Nutanix Enterprise AI dashboard, administrators utilize theImport Modelworkflow.

Workflow:In the "Import Model" wizard, there is a specific option to import fromHugging Face.1This requires the user to input theRepository ID(often referred to in generic UI options as the "URL" or path to the model, e.g., meta-llama/Llama-2-7b).

Why D is correct:This option allows the platform to pull the model weights and configuration files directly from the specified public repository.

Why others are incorrect:

Using CURL (A):NAI provides cURL code snippets toconsume(make inference requests to) a running model, not toimportit into the registry.

NGC Catalog (B):This is for NVIDIA-specific models, not general Hugging Face repositories.

Marketplace app (C):This typically refers to Nutanix Marketplace blueprints, not the specific action of importing a raw model weight set from Hugging Face.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

Nutanix Enterprise AI provides an inference gateway that is API-compatible with the OpenAI standard, allowing developers to use standard libraries (like the OpenAI Python SDK) to interact with models running on NAI. The supported endpoints that map to this standard include:

/v1/chat/completions: For chat-based text generation (supported).

/v1/embeddings: For generating vector embeddings (supported).

/v1/models: For listing the available models (supported).

The option/v1/systemsisnota supported or valid OpenAI-compatible endpoint. In the OpenAI API schema, "system" refers to a messagerole(e.g., {"role": "system", "content": "..."}), not a standalone URL endpoint. Therefore, this is the unsupported type in the list.

Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:

Nutanix Enterprise AI utilizes Role-Based Access Control (RBAC) to manage permissions.6

AI/ML Admin:This role has full privileges over the NAI platform, including the ability tocreate new users, assign roles (Admin vs.7User), manage global quotas, and view all endpoints and models across the system.

AI/ML User:This role is restricted to managing only theirownresources (models, endpoints, and keys) and cannot create other user accounts or view resources belonging to other users.