About Auditing Models¶
NVIDIA NeMo Auditor is released with early access availability and is subject to limited support and potential API changes in future releases.
NVIDIA NeMo Auditor audits LLMs by probing them with adversarial prompts and detecting failures such as jailbreaks, prompt injection, encoding bypasses, and unsafe output generation. It is powered by garak, NVIDIA's open-source LLM vulnerability scanner, and integrates with NeMo Platform so audits can target any model reachable through the Inference Gateway.
Typical Workflow¶
A typical audit looks like the following:
- Create an audit target for the model you want to test.
- Create an audit configuration that selects which garak probes and detectors to run, along with reporting settings.
- Run the audit and inspect the resulting JSONL, HTML, and hitlog reports.
The plugin exposes both synchronous and asynchronous Python entry points for each step.
Setup¶
Before you can run audits, you need a working NeMo Platform install with the auditor plugin enabled and a garak interpreter on disk.
- Follow Setup to install the platform and start local services.
- Install garak in a Python virtual environment. By default the plugin invokes
~/.auditor/.venv/bin/python -m garak; override the interpreter path withNEMO_AUDITOR_GARAK_PYTHONif you installed it elsewhere. - Configure at least one Inference Gateway provider so audits can route requests to the model under test.
Task Guides¶
-
Define the model under test — generator type, model identifier, and inference endpoint.
-
Choose probes, detectors, and reporting settings for the audit.
-
End-to-end walkthrough: create entities, run the audit in-process, read the report artifacts.
-
Reference for the
client.auditorSDK surface:configs,targets, andrun().
References¶
-
Field reference for
AuditConfigand itssystem,run,plugins, andreportingsub-models. -
Field reference for
AuditTarget(type,model,options). -
probe_spec,probe_tags, anddetector_specsyntax with worked examples. -
How
nmp_uri_specresolves a target's URI through a NeMo Platform provider.