dev-instructed-deception-NVIDIA-Nemotron-3-Super-120B-A12B-BF16-None
收藏资源简介:
该数据集是一个用于测试的基准数据集,包含216个测试样本。数据采用结构化格式,主要特征包括:模型标识(model)、LoRA适配器配置(lora)、样本索引(index)、多轮对话消息(messages)以及温度参数(temperature)和校验标识(canary)。其中messages字段为列表结构,每条消息包含内容(content)和角色(role)信息,表明数据集记录了完整的对话交互序列。该数据集适用于语言模型评估、对话系统测试、LoRA适配器性能分析等任务,通过控制温度参数可用于研究生成多样性对模型输出的影响。
This dataset is a benchmark dataset for testing, containing 216 test samples. The data is in a structured format, with key features including: model identifier (model), LoRA adapter configuration (lora), sample index (index), multi-turn dialogue messages (messages), temperature parameter (temperature), and verification identifier (canary). The messages field is a list structure, where each message contains content and role information, indicating that the dataset records complete dialogue interaction sequences. It is suitable for tasks such as language model evaluation, dialogue system testing, and LoRA adapter performance analysis, and by controlling the temperature parameter, it can be used to study the impact of generation diversity on model output.
- 数据集名称:dev-instructed-deception-NVIDIA-Nemotron-3-Super-120B-A12B-BF16-None
- 数据集提供方:aletheias-quest
- 配置:default
- 数据划分:仅包含 test 集
- 示例数量:216
- 总字节数:167,356 字节
- 下载大小:99,797 字节
- 数据集大小:167,356 字节
- 特征字段:
- model (string):模型名称
- lora (string):LoRA 配置
- index (int64):索引编号
- messages (list of objects):对话消息列表,每条消息包含:
- content (string):消息内容
- role (string):消息角色(如 user、assistant)
- temperature (float64):生成温度参数
- canary (string):金丝雀标记(用于测试或验证)
- 数据文件路径:data/test-*(位于数据集仓库根目录下)




