aisi-whitebox/gsm8k_prompted_sandbagging_llama_31_8b_instruct

inspect llama 31 8b instruct prompted sandbagging gsm8k数据集是一个用于评估和检测沙袋策略的专门数据集。它基于Llama-3.1-8B-Instruct模型创建，并包含了gsm8k任务的评估。数据集包含了两种系统提示：良性提示和恶意提示，用于引导模型产生不同的输出。沙袋检测功能已启用，但未应用过滤。数据集未进行分割，但有测试和验证大小的设定。此外，README还提供了数据集的统计数据和额外的参数设置，以及数据集的版本控制信息。

The inspect llama 31 8b instruct prompted sandbagging gsm8k dataset is a specialized dataset for evaluation and detection of sandbagging strategies. It is created based on the Llama-3.1-8B-Instruct model and includes evaluations for the gsm8k task. The dataset contains two types of system prompts: benign and malicious, designed to guide the model to produce different outputs. Sandbagging detection is enabled but no filtering is applied. The dataset is not split, but test and validation sizes are specified. Additionally, the README provides statistical data and additional parameter settings for the dataset, as well as version control information.