dev-varied-deception-Qwen3.5-27B-a-mo-qwen3.5-27b-5
收藏资源简介:
该数据集是一个用于测试或评估的结构化数据集,包含400个样本,仅设有测试分割。每个样本由多个字段构成:model(字符串类型,可能表示使用的基座模型)、lora(字符串类型,可能表示应用的LoRA适配器)、index(整型,样本索引)、messages(列表类型,包含多轮对话消息,每条消息有content(内容)、reasoning(推理过程)和role(角色)三个子字段)、temperature(浮点型,可能表示生成温度参数)以及canary(字符串类型,用途未明确说明)。数据格式表明其可能专为大型语言模型(LLM)的对话生成、推理能力评估或不同模型/配置(如结合LoRA)的对比测试而设计。数据集总大小约为3.8 MB。
This is a structured dataset designed for testing and evaluation purposes, containing 400 samples with only a test split. Each sample comprises multiple fields: 1. `model`: a string type field, possibly indicating the base model used; 2. `lora`: a string type field, possibly indicating the applied LoRA adapter; 3. `index`: an integer type field representing the sample index; 4. `messages`: a list containing multi-turn conversation messages, where each message includes three sub-fields: `content` (the message content), `reasoning` (the reasoning process), and `role` (the speaker role); 5. `temperature`: a float type field, possibly representing the generation temperature parameter; 6. `canary`: a string type field with an unspecified specific purpose. The data format indicates that this dataset may be specifically tailored for dialogue generation and reasoning capability evaluation of large language models (LLMs), or comparative testing across different models or configurations (e.g., configurations incorporating LoRA). The total size of the dataset is approximately 3.8 MB.
- 数据集名称: dev-varied-deception-Qwen3.5-27B-a-mo-qwen3.5-27b-5
- 数据集地址: https://huggingface.co/datasets/aletheias-quest/dev-varied-deception-Qwen3.5-27B-a-mo-qwen3.5-27b-5
- 配置: 默认配置(
default),包含一个测试集(test)分片,数据文件路径为data/test-* - 数据集大小: 下载大小为 3,660,029 字节,数据集总大小为 3,792,837 字节
- 样本数量: 测试集包含 400 个样本
- 特征字段:
model(字符串): 模型名称lora(字符串): LoRA 配置index(整型): 索引编号messages(列表): 消息列表,每条消息包含以下字段:content(字符串): 消息内容reasoning(字符串): 推理过程role(字符串): 角色(如用户或助手)
temperature(浮点数): 温度参数canary(字符串): 金丝雀标记(用于检测数据泄漏)




