Subset metrics by evidence source:
Pure-text (Plain-text): samples=112, accuracy=42.86%
Figure: samples=100, accuracy=37.00%
Generalized-text (Layout): samples=65, accuracy=36.92%
Table: samples=65, accuracy=41.54%
Chart: samples=64, accuracy=42.19%
Subset metrics by evidence pages length:
no_pages: samples=78, accuracy=48.72%
single_page: samples=159, accuracy=50.31%
multiple_pages: samples=108… See the full description on the dataset page:
https://huggingface.co/datasets/happy8825/MMLongBench_var11_prompt_enhanced_0111.
