Welcome to MMTrustEval

• MMTrustEval (MMTE) is a toolbox developed for the benchmark, MultiTrust (Benchmarking Trustworthiness of Multimodal Large Language Models, paper)
• It provides a universal and scalable infrastructure for evaluating MLLM trustworthiness and facilitating future research.
• Different MLLMs are integrated into a unified interface to conduct standardized inference.
• Tasks are modularized by separting data, inference, and evaluation metrics to encourage tool reuse and easy updates for new tasks to be added.