dataset_info:
features:
- name: instruction
dtype: string - name: demonstration
dtype: string
splits: - name: train
num_bytes: 34540085.04363476
num_examples: 65499
download_size: 0
dataset_size: 34540085.04363476
license: mit
pretty_name: Helpful Raw Anthropic
tags: - human-feedback
Dataset Card for "helpful-raw-anthropic"
This is a dataset derived from Anthropic's HH-RLHF data of instructions and model-generated demonstrations. We combined training splits from the following two subsets:
helpful-basehelpful-online
To convert the multi-turn dialogues into (instruction, demonstration) pairs, just the first response from the Assistant was included. This heuristic captures the most obvious answers, but overlooks more complex questions where multiple turns were required to get a helpful response. Some additional filtering is likely required (e.g. defining a minimun length or computing ROUGE-L scores with the instruction/demonstration).