Layer-wise downstream results for 17 music foundation models. Each cell shows
the best layer's score, its index, and the full layer profile as a sparkline.
Click any column header to sort; family headers (▸) expand companion metrics.
Results as of 2026-08-04
Rank = mean of the model’s per-task ranks (1 = best;
p ranked on fewer tasks)
Params = evaluated encoder (audio tower for two-tower; decoder for AR)
+ beyond the paper’s 12 models
† metric caveat (hover) — not evaluated (coverage)
dot = best layer