Galileo Guardrail Metrics

Understand Galileo's Guardrail Metrics in LLM Studio

Galileo has built a menu of Guardrail Metrics to help you evaluate and observe your generative AI applications. These metrics are tailored to your use case and are designed to help you ensure your application quality and behavior. The Scorer definition for each metric is listed immediately below.

Galileo's Guardrail Metrics are a combination of industry-standard metrics and an outcome of Galileo's in-house ML Research Team.

Output Quality Metrics

• Correctness Scorers.factuality

• Uncertainty Added by default

• BLEU Added by default

• ROUGE-1 Added by default

RAG Quality Metrics

• Context Adherence

  • Context Adherence Basic: Scorers.context_adherence_basic

  • Context Adherence Plus: Scorers.context_adherence_plus

• Completeness

  • Completeness Basic: Scorers.completeness_basic

  • Completeness Plus: Scorers.completeness_plus

• Chunk Attribution

  • Chunk Attribution Basic: Scorers.chunk_attribution_utilization_basic

  • Chunk Attribution Plus: Scorers.chunk_attribution_utilization_plus

• Chunk Utilization

  • Chunk Utilization Basic: Scorers.chunk_attribution_utilization_basic

  • Chunk Utilization Plus: Scorers.chunk_attribution_utilization_plus

• Context Relevance Scorers.context_relevance

Input Quality Metrics

• Prompt Perplexity Scorers.prompt_perplexity

Safety Metrics

• Input & Output PII Scorers.pii

• Input & Output Tone Scorers.tone

• Input & Output Toxicity Scorers.toxicity

• Input & Output Sexism Scorers.sexist

• Prompt Injection

  Scorers.prompt_injection

Looking for something more specific? You can always add your own custom metric.