exploring-llm-clusters

To analyze LLM usage patterns, you group similar traces and generations into clusters to reveal common behaviors. This enables efficient pattern discovery and allows you to compare cluster sizes and centroid metrics across clustering runs.

The best way to compare LLM cluster behavior is to list and retrieve clustering runs, inspecting cluster size, centroid metrics, and descriptive titles. This allows you to evaluate differences and identify prominent usage patterns across your AI traffic.

Yes, you can compute cost and latency metrics for individual LLM traces within a cluster. By drilling down into specific generations, you can inspect granular behavior and evaluate the performance of each item.

To identify common usage patterns in AI traffic, apply clustering to your LLM traces and generations. This groups similar interactions, enabling you to drive investigations by inspecting top clusters and evaluating related cost and latency metrics.

Yes, you need to load a specific clustering run to inspect LLM analytics clusters. Loading the latest run allows you to retrieve the cluster data, inspect top clusters, and identify patterns to drive your investigations.