TeleAI-AI-Flow/InformationCapacity

信息容量（Information Capacity）是一种基于文本压缩性能与计算复杂度相关性的大型语言模型（LLM）效率评估方法。该方法通过比较不同大小的模型在一系列任务中的表现，来预测模型性能。它还考虑了标记器效率对模型评估的影响。五个异构数据集被用于评估LLM的压缩效率。

Information Capacity is an efficiency evaluation method for large language models (LLM) based on the correlation between text compression performance and computational complexity. This method predicts model performance by comparing the performance of models of varying sizes across a series of tasks. It also considers the influence of tokenizer efficiency on model evaluation. Five heterogeneous datasets are used to assess the compression efficiency of LLMs.