What AI says about WildBench

WildBench is an open-source benchmarking framework for evaluating large language models on real-user, challenging tasks in the wild. It provides a leaderboard and a dataset evaluation framework with scripts to run batch evaluations and compute scores (WB-Score) to compare model performance. It supports integration with HuggingFace and vLLM, with documentation and tooling to add new models to the benchmark.