Fiddler Auditor: Open-source tool evaluates the robustness of large language models

Fiddler Auditor is an open-source tool designed to evaluate the robustness of Large Language Models (LLMs) and Natural Language Processing (NLP) models. LLMs can sometimes produce unwarranted content, potentially create hostile responses, and may disclose confidential information they were trained with, regardless of whether they were explicitly asked to do so. The tool uses adversarial examples, out-of-distribution inputs, and linguistic variations to help developers and researchers identify potential weaknesses and improve the performance of their … More →

The post …

confidential don't miss fiddler github information language language models large llms may natural language natural language processing nlp open source robustness software tool unwarranted