Reported on 25 benchmarks across 10 tasks · 4 papers · 11 SOTA
Note: results are matched by exact model name. Different papers may use the same name for different model variants.