✨ 简介

There isn't a single correct answer to this question. With the rapid pace of AI model development, new open-source models are released every week, often claiming to outperform previous versions. The best model for your needs depends largely on your GPU capacity and the type of data you're working with. It's a good idea to explore newer, widely accepted models with strong general capabilities. You can refer to 此列表 for available open-source models, their release dates, and fine-tuned variants.

NaN stands for "Not a Number." In ragas evaluation results, NaN can appear for two main reasons:

• JSON解析问题： 模型的输出无法被JSON解析。ragas要求模型输出JSON兼容的响应，因为所有提示都是使用Pydantic结构化的。这确保了LLM输出的高效解析。
• 不适合评分的情况： 样本中的某些情况可能不适合进行评分。例如，对"我不知道"这类回答进行真实性评分可能就不太合适。

The best way is to trace and log your evaluation, then inspect the results using LLM traces. You can follow a detailed example of this process here.