license: mit
task_categories:
- text-generation
language: - en
tags: - human preference alignment
size_categories: - 100K<n<1M
Dataset Card for HH-RLHF-MetaAligner-Data
Dataset Summary
This repository provides access to 2 subsets of the MetaAligner data built from HH-RLHF:
- The equal-preference subset is included in: equal_train.json and equal_val.json.
- The preference subset is included in: preference_train.json, preference_val.json, and preference_test.json.
Disclaimer: As the dialogue data comes from HH-RLHF, the data contain content that may be offensive or upsetting. Topics include, but are not limited to, discriminatory
language and discussions of abuse, violence, self-harm, exploitation, and other potentially upsetting subject matter. Please only engage with the data following your own
personal risk tolerance. The data are intended for research purposes, especially research that can make models less harmful.
Usage
Each of the above datasets is located in a separate sub-directory. To load an individual subset, use the data_dir argument of the load_dataset() function as follows:
from datasets import load_dataset
# Load the preference subset
data_files = {"train": "preference_train.json", "validation": "preference_val.json", "test": "preference_test.json"}
preference_dataset = load_dataset('MetaAligner/HH-RLHF-MetaAligner-Data', data_files=data_files)
# Load the equal-preference subset
data_files = {"train": "equal_train.json", "validation": "equal_val.json"}
equal_dataset = load_dataset('MetaAligner/HH-RLHF-MetaAligner-Data', data_files=data_files)