The safety dashbaord allows you to monitor your model’s performance and reliability with a comprehensive dashboard tracking key metrics, critical summaries, and live data insights

​

Key Metrics Overview

1. Prompt Safety Violation Rate: % of times user sends unsafe requests or instructions.
2. Response Safety Violation Rate: % of times the AI sends unsafe response to the user
3. False Refusal Rate: % of times the AI refuses to follow a safe request or instruction

The colours correspond to the following values:

Colors: Red —> >5% Orange —> 1% to 5% Green —> < 1%

Response Red —> >3% Orange —> 1% to 3% Green —> < 1%

False refusal rate: Red —> >2% Orange —> 0.5-2% Green —> < 0.5

​

Critical Summaries

The Critical Summaries gives you insights about:

1. Total Queries vs Flagged Queries graph over time
2. Risk Levels
3. Flagged Categories

​

Live Data

The Live Data table gives you data about

1. ID
2. Conversation Prefix
3. Assistant Response
4. Judge Output
5. Feedback
6. Categories