Reported on 35 benchmarks across 14 tasks · 4 papers · 16 SOTA
Note: results are matched by exact model name. Different papers may use the same name for different model variants.