Reported on 24 benchmarks across 8 tasks · 3 papers · 22 SOTA
Note: results are matched by exact model name. Different papers may use the same name for different model variants.