Loading cluster trends data...
Cluster Fraction by Layer Depth
Each point is a (model, layer, cluster). Y-axis shows fraction of
heads at that layer belonging to each cluster.
Cluster Fraction by Model Size
Each point is a (model, cluster). X-axis is parameter count (log
scale).
Cluster Entropy by Layer Depth
Shannon entropy of cluster distribution at each layer. Higher = more
diverse clusters, lower = dominated by one cluster.