Skip to content

Dataset · MaziyarPanahi

180 thousand evaluations and decisions for AI policy verification

MaziyarPanahi/Jev-Policy-Verification-180K

Tabular and text corpus in English for RLHF, response evaluation, and model policy compliance checking.

What's inside

It contains 180,000 English records, each linking a ground truth label to a jev-1.13.0 model prediction, confidence score, structured response, and provenance data. The data covers quality evaluation, preferences, response helpfulness, and adherence to specified principles (there is no medical data in the set). The dataset is published in flat parquet format.

What you can do with it

  • Training reward models and aligning LLMs via RLHF methods
  • Training classifiers and rankers to check AI responses for rule compliance
Tasks
Text classification, Text ranking, Reinforcement learning
Languages
English
Size
100K–1M rows
Downloads this month
258
Likes
13

How to load it

from datasets import load_dataset

ds = load_dataset("MaziyarPanahi/Jev-Policy-Verification-180K")