Add Convence/ParseEmbed as an official benchmark on the Hub (If possible)

🤗Datasets
1

Hi Hugging Face team,

I’d like to register Convence/ParseEmbed as an official benchmark on the Hub.

Dataset: Convence/ParseEmbed · Datasets at Hugging Face

It includes a root eval.yaml with:

  • name: ParseEmbed
  • evaluation_framework: mteb
  • tasks: mean, text_formatting, table
  • config: parse-embed

ParseEmbed is a retrieval benchmark for embedding models. It tests whether models preserve parse-sensitive meaning under hard negatives, including semantic scope, formatting-sensitive text, and table grounding.

The dataset card documents the purpose, files, task IDs, and usage. The dataset loads with datasets using the parse-embed config and the task splits.

Could you please add it to the official benchmark allow-list?

Thanks! if something is wrong let me know

1 Like