Reported on 60 benchmarks across 9 tasks · 4 papers · 8 SOTA
Note: results are matched by exact model name. Different papers may use the same name for different model variants.