VOOZH about

URL: https://huggingface.co/datasets/qualifire/grounding-benchmark

⇱ qualifire/grounding-benchmark · Datasets at Hugging Face

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

The Grounding Claims Dataset is a multi-domain dataset for evaluating whether a natural language claim is grounded (i.e., supported or entailed) by a document. The dataset is organized into four subsets, each requiring different types of reasoning:

  • general (1500 examples): Broad, everyday reasoning
  • logical (1000 examples): Logical consistency and inference
  • time_and_dates (100 examples): Temporal reasoning
  • prices_and_math (100 examples): Numerical and mathematical reasoning

Each entry consists of:

  • doc: A short context or passage
  • claim: A natural language statement to verify against the doc
  • label: A binary label indicating whether the claim is grounded in the document (1 for grounded, 0 for ungrounded)
  • dataset: The source subset name (e.g., "general")

📌 Features

Feature Type Description
doc string The document or passage providing the context
claim string A statement to verify against the document
label string grounded or ungrounded
dataset string The domain/subset the instance belongs to

📊 Usage

This dataset can be used to train and evaluate models on factual verification, natural language inference (NLI), and claim grounding tasks across multiple domains.

🏷️ Labels

  • grounded — The claim is grounded in the document.
  • ungrounded — The claim is ungrounded or contradicted by the document.
Downloads last month
16

Models trained or fine-tuned on qualifire/grounding-benchmark

Collection including qualifire/grounding-benchmark