Proposed: B2 + B4 w/ multimodal fusion

Reported on 2 benchmarks across 1 task · 1 paper

Note: results are matched by exact model name. Different papers may use the same name for different model variants.