Skip to content

Dataset · MaziyarPanahi

Claim verification and hallucination detection in RAG

MaziyarPanahi/Jev-RAG-Claim-Verification-130K

130 thousand examples of response source verification with annotation of hallucination types and errors.

What's inside

The dataset contains 130,000 English records linking reference labels from sources with predictions from the jev-1.13.0 model. Individual claims, overall context grounding of the response, hallucination types, and formatting defects are annotated in tabular and text formats. Additionally, 21,181 cases of discrepancy between the teacher model and the reference are preserved.

What you can do with it

  • Training classifiers for hallucination detection in RAG systems
  • Testing and evaluating the quality of judge models (LLM-as-a-Judge)
  • Analyzing discrepancies between reference answers and model generation
Tasks
Text classification
Languages
English
Size
100K–1M rows
Downloads this month
708
Likes
43

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("MaziyarPanahi/Jev-RAG-Claim-Verification-130K")