Request to add real5-omnidocbench framework and Benchmark allow list entry

🤗Datasets
1

Hi Hugging Face team,

We would like to add a new Hub Benchmark evaluation framework and register our dataset as a native Hub Benchmark.

Requested framework identifier:

real5-omnidocbench

Dataset:

Reason:
Real5-OmniDocBench evaluates document parsing robustness under five real-world acquisition scenarios: Scanning, Warping, Screen-Photography, Illumination, and Skew. Existing framework identifiers such as olmocr-bench, mdpbench, and parsebench do not identify this benchmark directly.

Current status:

  • The dataset has a root eval.yaml.
  • The eval.yaml uses evaluation_framework: real5-omnidocbench.
  • The dataset currently has the generic benchmark tag.
  • It does not yet show benchmark:eval-yaml or benchmark:official.

Proposed enum entry:

“real5-omnidocbench”: {
name: “real5-omnidocbench”,
description:
“Real5-OmniDocBench is a real-world document parsing robustness benchmark covering five physical acquisition scenarios: Scanning, Warping, Screen-Photography, Illumination, and Skew.”,
url: " PaddlePaddle/Real5-OmniDocBench · Datasets at Hugging Face ",
}

Could you help us with:

  1. Adding real5-omnidocbench to the supported evaluation framework enum.
  2. Adding PaddlePaddle/Real5-OmniDocBench to the Benchmark allow-list so the native Benchmark Leaderboard UI can appear.

Thanks.

1 Like