Reported on 30 benchmarks across 5 tasks · 1 paper
Note: results are matched by exact model name. Different papers may use the same name for different model variants.