yusufbaykaloglu/Human-Like-DPO-Dataset-TR
收藏Hugging Face2025-11-06 更新2025-11-15 收录
下载链接：
https://hf-mirror.com/datasets/yusufbaykaloglu/Human-Like-DPO-Dataset-TR
下载链接
链接失效反馈官方服务：
资源简介：
Human-Like-DPO-Dataset-TR是一个土耳其语数据集，旨在提高大型语言模型中的对话自然度和流畅性。它包含了10884个示例，涵盖256个主题，适用于Direct Preference Optimization (DPO)和Supervised Fine-Tuning (SFT)方法。每个示例由三个字段组成：prompt（日常对话问题）、chosen（人类似、自然的回答）和rejected（官方、传统的AI回答）。
Human-Like-DPO-Dataset-TR is a Turkish dataset designed to enhance the naturalness and fluency of dialogues in large language models. It contains 10,884 examples covering 256 topics and is optimized for Direct Preference Optimization (DPO) and Supervised Fine-Tuning (SFT) methods. Each example consists of three fields: prompt (daily conversation questions), chosen (human-like, natural responses), and rejected (official, traditional AI responses).
提供机构：
yusufbaykaloglu
