aisi-whitebox/uriah_dataset_generation_claude_3_7_sonnet_20250219_mmlu_0_shot_cot

该数据集使用anthropic/claude-3-7-sonnet-20250219模型生成，于2025-06-26创建，包含mmlu_0_shot_cot任务。数据集未进行分割，但有测试集和验证集的大小设置，分别为20%和50%，使用随机种子42。其他参数包括数据集大小限制为10，错误容忍率为0.2，训练1个周期，最大连接数为32，token数量限制为100000。数据集的生成目的是用于评估，涉及欺骗性和安全性，特别是针对沙袋策略的检测和过滤。

This dataset was generated using the anthropic/claude-3-7-sonnet-20250219 model and created on 2025-06-26, containing the task mmlu_0_shot_cot. The dataset is not split, but the test set and validation set sizes are set to 20% and 50% respectively, with a random seed of 42. Other parameters include a dataset size limit of 10, an error tolerance rate of 0.2, training for 1 epoch, a maximum of 32 connections, and a token limit of 100000. The dataset is intended for evaluation purposes, involving deception and safety, particularly focusing on sandbagging detection and filtering.