Skip to content

Evals

Optional gllm-evals integration surface for guardrail-owned metrics.

BiasMetric(dm_invoker, threshold=0.5, strict_mode=False)

Bases: _TrustSecurityMetric

Guardrail response-bias metric where lower violation probability is better.

Attributes:

Name Type Description
name str

Canonical metric name, guardrail_bias.

ToxicityMetric(dm_invoker, threshold=0.5, strict_mode=False)

Bases: _TrustSecurityMetric

Guardrail response-toxicity metric where lower violation probability is better.

Attributes:

Name Type Description
name str

Canonical metric name, guardrail_toxicity.