kivi-1.5b-veri
收藏资源简介:
Kivi Elektronik Sentetik Eğitim ve Test Veri Seti(Kivi Elektronik 合成训练与测试数据集)是一个土耳其语合成文本生成数据集,采用 MIT 许可证,语言标记为 tr,标签包括 synthetic-data、rag 和 agent。该数据集在 Tech Istanbul 人工智能训练营毕业项目范围内创建,目的是微调 Qwen/Qwen2.5-1.5B-Instruct 模型。数据集共有 406 个样本,其中 366 个用于训练,40 个用于测试。样本类型包括约 99 个身份与风格对话、约 168 个基于指南的 RAG 示例、约 120 个单轮 ReAct 工具调用示例,以及约 19 个两轮观察与最终回答示例。数据内容围绕虚构的 Kivi Elektronik 客户服务场景组织,适用于文本生成、RAG、Agent/ReAct 工具调用等训练与测试任务。
Kivi Elektronik Synthetic Training and Test Dataset is a Turkish synthetic text generation dataset under the MIT license, with language tag tr and tags including synthetic-data, rag, and agent. It was created within the scope of the Tech Istanbul Artificial Intelligence Bootcamp graduation project, with the purpose of fine-tuning the Qwen/Qwen2.5-1.5B-Instruct model. The dataset contains 406 samples, with 366 used for training and 40 for testing. Sample types include about 99 identity and style dialogues, about 168 guideline-based RAG examples, about 120 single-turn ReAct tool-calling examples, and about 19 two-turn observation and final answer examples. The content is organized around a fictional Kivi Elektronik customer service scenario and is suitable for training and testing tasks such as text generation, RAG, and Agent/ReAct tool calling.





