lex-eval

https://github.com/LegionIO/lex-eval

Provides LLM-as-judge and code-based evaluators for scoring LLM outputs, with built-in templates for hallucination, relevance, and toxicity detection.