The standard for evaluating text is human labeling. However, human evaluation is often impractical at scale. Evaluating the performance of LLM applications is increasingly handled by using a separate evaluation LLM (LLM as a judge 👩🏾⚖️). LLM evaluation is a great starting point for understanding where an LLM application goes wrong. This demo covers running an LLM evaluation using Arize Phoenix, including evals with explanations for Q&A correctness and hallucinations. The Arize Phoenix LLM Evals open source library is designed for simple, fast, and accurate LLM-based evaluations. It leverages a variety of LLM evaluation metrics and tracing.
Dive into the 📓notebook.