Reported on 28 benchmarks across 12 tasks · 6 papers · 17 SOTA
Note: results are matched by exact model name. Different papers may use the same name for different model variants.