Reported on 30 benchmarks across 11 tasks · 7 papers · 4 SOTA
Note: results are matched by exact model name. Different papers may use the same name for different model variants.