eekay/gemma-2b-it-noised-np0.15-emb-lion-numbers

该数据集是通过在Gemma-2b-it模型上应用噪声注入生成的，具体在blocks.14.hook_resid_post钩子点使用add_bias_hook_fn函数。数据集包含30,000个示例，每个示例由3到10个数字组成，数字值范围从0到999。每个示例生成10个答案，答案数字最大位数为3。数据集设计用于训练或评估模型在噪声环境下的数字序列生成能力，支持批处理大小为64，最大新令牌数为96。数据集以MIT许可证发布，语言为英语。

This dataset is generated by applying noise injection to the Gemma-2b-it model, specifically at the blocks.14.hook_resid_post hook point using the add_bias_hook_fn function. It consists of 30,000 examples, each comprising 3 to 10 numbers with values ranging from 0 to 999. Each example generates 10 answers, with a maximum digit count of 3. The dataset is designed for training or evaluating model performance in generating number sequences under noisy conditions, with a batch size of 64 and a maximum of 96 new tokens. It is released under the MIT license and is in English.