Facebook's Factual Reasoning dataset contains training data for the 'Learning to Reason for Factuality' paper. The dataset is hosted on Hugging Face and was last updated on September 25, 2025. The associated scalable VeriScore implementation is available on GitHub.
Use Cases
- Training models for factual reasoning based on the described training data.
- Evaluating model factuality using the scalable VeriScore implementation mentioned.
- Benchmarking AI systems on tasks related to truthfulness and verification.
Strengths
- Training data is directly linked to a published research paper ('Learning to Reason for Factuality').
- Associated scalable VeriScore implementation is publicly available on GitHub.
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
Provenance
- Source
- facebook
- Freshness
- Last updated 2025-09-25 23:28:54; freshness should be verified.