DeepEval

Open-source LLM evaluation framework. 14+ metrics (G-Eval, hallucination, faithfulness, toxicity), pytest-style assertions, and synthetic dataset generation. Unit-test LLM outputs the way you unit-test code.