ResearchBenchmarksData-mixture learning on 17 domains (proxy-train)Follow7,436Peak Memory (MB)BROS7,193.5648,830.00710,466.4512,102.893May 11, 2026Evaluation ResultsMethodMethodLinksPeak Memory (MB)Memory vs. MA-SOBA RatioPer-Step Time (s)BROSr/n=0.25r/n=0.252026.057,43655.10.301DoReMi2026.059,447700.149ZOFO2026.0510,047.174.40.556Penalty2026.0510,190.575.50.272FdeHBO2026.0512,896.395.60.26MA-SOBA2026.0513,496.91000.278