Reported on 64 benchmarks across 32 tasks · 10 papers · 26 SOTA
Note: results are matched by exact model name. Different papers may use the same name for different model variants.