ryota39/

open_preference_v0.1

public RLHF dataset in Japanese Quality of Japanese text is somewhat low arise from the combination of synthetic generated text and machine translation API

	details




	

dataset_name num_samples how_to_make_data license

llm-jp/hh-rlhf-12k-ja 12k Anthropic社が開発初期段階のClaude-52Bを用いてユーザー入力と応答文を生成, chosen/rejectedは手作業で判定, 左記データセットをdeeplで翻訳 MIT

cyberagent/chatbot-arena-ja-calm2-7b-chat-experimental 29k ユーザー入力:chatbot-arenaのユーザー入力, CC-BY 4.0のユーザ入力… See the full description on the dataset page: https://huggingface.co/datasets/ryota39/open_preference_v0.1.