saturnMars/hyperprobe-dataset-analogy
收藏官方服务:
资源简介:
hyperprobe-dataset是一个用于解码大型语言模型(LLM)向量空间的人类可解释特征的合成数据集,包含事实和语言类比语料库以及基于SQuAD的问答任务语料库。
The hyperprobe-dataset is a synthetic dataset for decoding human-interpretable features from the vector space of large language models (LLM), including corpora of factual and linguistic analogies and SQuAD-based question-answering tasks.
提供机构:
saturnMars


