name: olmOCR-bench
description: olmOCR-bench is a benchmark to evaluate how well VLLMs can parse PDF and scanned documents into clean, semantic markdown/HTML.
evaluation_framework: olmocr-bench

tasks:
  - id: overall
    config: default
    split: overall     
  - id: arxiv_math
    config: default
    split: arxiv_math
  - id: headers_footers
    config: default
    split: headers_footers
  - id: long_tiny_text
    config: default
    split: long_tiny_text
  - id: multi_column
    config: default
    split: multi_column
  - id: old_scans
    config: default
    split: old_scans
  - id: old_scans_math
    config: default
    split: old_scans_math        
  - id: table_tests
    config: default
    split: table_tests       
  - id: baseline
    config: default
    split: baseline