Managing Audit Targets#

When you run an audit job, you specify a separate audit target and an audit configuration for the job.

By using separate target and configurations, you can create a configuration one time and reuse it multiple times with different targets. For example, you can create multiple targets and jobs with the same configuration to compare model.

Accessing the Microservice#

Set the AUDITOR_BASE_URL according to how you deployed the microservice.

Deployment Technology

Variable Value

Quickstart with Docker Compose

export AUDITOR_BASE_URL=http://localhost:8080

Demo Cluster with minikube

export AUDITOR_BASE_URL=http://nemo.test

Installed individually using Helm

If you needed to port forward the service, then export AUDITOR_BASE_URL=http://localhost:5000.

If you have an ingress controller, contact your Kubernetes administrator.

Task Guides#

Perform common audit target tasks.

Create a Basic Target

Create a new target

Basic Audit Target
Create a build.nvidia.com Target

Create a target for accessing models from integrate.api.nvidia.com

Build.nvidia.com Model Target
Create a NeMo NIM Proxy Target

Create a target for accessing models from NeMo NIM Proxy

NeMo NIM Proxy Target
Create a Locally-Deployed NIM Target

Create a target for accessing models from a locally-deployed NIM for LLMs microservice

Locally-Deployed NIM Microservice Target
Create a Target for OpenAI

Create a target for accessing models from api.openai.com

OpenAI Model Target
List Targets

List all audit targets

List Audit Targets
Update and Delete Targets

Manage audit targets

Updating and Deleting Audit Targets

References#

Review detailed specifications for creating audit configurations.

Target Schema

Reference documentation for the JSON schema used to define audit configurations

Schema for Audit Targets