Add olmOCR-bench old_scans experiment for PaddleOCR-VL-1.6#25

Merged

davanstrien merged 7 commits into

mainfrom

add-olmocr-bench-oldscans

Jun 27, 2026

Commits on Jun 27, 2026

Add olmOCR-bench old_scans experiment for PaddleOCR-VL-1.6

davanstrien
and
claude
committed
Tighten README + convert docstring to a factual spec

davanstrien
and
claude
committed
README: record greedy decoding + eyeball spot-check in status

davanstrien
and
claude
committed
Add samples.html (scan vs output, CJK highlighted) + generator

davanstrien
and
claude
committed
Address review: published number is the OLDER PaddleOCR-VL, not 1.6

davanstrien
and
claude
committed
De-duplicate PDF handling + parameterize version

davanstrien
and
claude
committed
Fold in v1 anchor: harness reproduces published 37.8 (got 38.2)

davanstrien
and
claude
committed