In addition to running evals, Athina provides advanced analytics and monitoring. It allows you to track usage patterns, monitor conversations, and gain insights from the data generated by your models. By leveraging these features, you can make data-driven decisions, optimize your models, and improve the overall user experience.

​
Logging Metrics

Inference analytics jobs will run once per day. We store data at the daily level, so we have granular “per-day” analytics, but not per-hour. You can filter inference analytics by the following fields to explore usage for specific segments:
  • date
  • environment
  • prompt_slug
  • customer_id
  • topic
  • language_model_id
You can explore your analytics in the Analytics Dashboard page or on the Observe page

​
View evaluation metrics

Evaluation analytics jobs run every hour. You can filter evaluation analytics by the following fields to explore performance for specific segments:
  • environment
  • prompt_slug
  • customer_id
  • topic
  • language_model_id
Currently, you can view the following evaluation metrics for each configured evaluation:
  • avg. evaluation score per day
  • percentile distributions of eval metrics
Head to your Analytics dashboard to view your model’s evaluation metrics.