DeepEval
Open-source framework of tests for model applications.
Обзор
DeepEval provides assertions for model outputs - answer relevancy, faithfulness, hallucination, contextual precision - and runs them as tests in a suite. It solves the problem that evaluation is usually a notebook somebody ran once, so nothing catches a regression when the prompt or the model changes. It exists because treating quality metrics as tests is the only way they get run on every change.
Граф технологий
Как DeepEval связана с остальным стеком.
Применяется к
- RAG technique Techniques DeepEval used for RAG High ✓ source
Инструменты
ПО, которое реализует DeepEval или интегрируется с ним.