Reported on 24 benchmarks across 7 tasks · 2 papers · 24 SOTA
Note: results are matched by exact model name. Different papers may use the same name for different model variants.