PertEval: Unveiling Real Knowledge Capacity of LLMs with Knowledge-Invariant Perturbations

Open in new window