DeepEval
DeepEval 是把软件测试那套写法搬到 LLM 上的开源评测框架：用 pytest 的风格给模型输出写断言，覆盖答案相关性、忠实度、幻觉、检索质量这些维度，跑起来像跑单测一样出通过率报告。评估指标由模型自己打分，可以接自己的模型，也能在本地跑，适合要把提示词和 RAG 改动作回归验证的团队。 🔗 文档: https://docs.confident-ai.com | GitHub Stars: 19k+
DeepEval is an open-source evaluation framework that borrows the shape of software testing: write pytest-style assertions against model output covering answer relevancy, faithfulness, hallucination and retrieval quality, then run it like a unit test suite and get a pass rate. Metrics are model-graded and can point at your own model or run locally, which suits teams regression-testing prompt and RAG changes. 🔗 Docs: https://docs.confident-ai.com | GitHub Stars: 19k+
📋 Product Info
🎯 Why This Tool
DeepEval belongs to the Developer Tools category on hedirbase. Tagged with: LLM评测,单元测试,开源,RAG评估.
DeepEval is an open-source evaluation framework that borrows the shape of software testing: write pytest-style assertions against model output covering answer relevancy, faithfulness, hallucination an