Reported on 28 benchmarks across 3 tasks · 2 papers · 25 SOTA
Note: results are matched by exact model name. Different papers may use the same name for different model variants.