Benchmark Results
A synthetic benchmark for measuring long-array extraction from documents:
the task of pulling a single long, repeating list of structured records out of a
PDF, completely and in order, when that list runs to hundreds or thousands of
rows and spills across many pages.
Most document-extraction benchmarks test header fields (a handful of scalars
per document). This benchmark isolates the failure mode that shows up at scale —
recall… See the full description on the dataset page:
https://huggingface.co/datasets/Extend-AI/LongArray-Extract.
