This repository contains the test splits for the MAPLE benchmark introduced in the paper MAPLE: Modality-Aware Post-training and Learning Ecosystem (
https://arxiv.org/pdf/2602.11596). The benchmark is designed for modality-aware multimodal evaluation under different required-signal settings, where each sample is annotated with the minimal modality subset needed to solve the task.
MAPLE-bench evaluates multimodal reasoning… See the full description on the dataset page:
https://huggingface.co/datasets/lihVerma/MAPLE-bench.
