Mobile-MMLU: A Mobile Intelligence Language Understanding Benchmark
Sondos Mahmoud Bsharat, Mukul Ranjan, Aidar Myrzakhan, Jiacheng Liu, Bowei Guo, Shengkun Tang, Zhuang Liu, Yuanzhi Li, Zhiqiang Shen
TL;DR
This paper introduces Mobile-MMLU and Mobile-MMLU-Pro, a pair of mobile-centric benchmarks designed to evaluate on-device language understanding under resource constraints. The authors propose a four-stage data construction pipeline and human–AI collaboration to ensure relevance, quality, and reduced bias, along with a more challenging Mobile-MMLU-Pro variant generated via multi-model consistency and rejection sampling. Empirical results across 1B–9B models show Mobile-MMLU better differentiates mobile-capable models than traditional desktop benchmarks, with notable variance among smaller models and a significant impact of answer ordering. The work emphasizes privacy, on-device processing, and personalized adaptation, offering a standardized framework to advance mobile-optimized LLMs and highlighting the need for mobile-aware benchmarks in real-world deployment. Overall, Mobile-MMLU provides a realistic, discriminative, and privacy-conscious benchmark suite that drives development of efficient, user-centric language technologies for mobile platforms.
Abstract
Rapid advancements in large language models (LLMs) have increased interest in deploying them on mobile devices for on-device AI applications. Mobile users interact differently with LLMs compared to desktop users, creating unique expectations and data biases. Current benchmark datasets primarily target at server and desktop environments, and there is a notable lack of extensive datasets specifically designed for mobile contexts. Additionally, mobile devices face strict limitations in storage and computing resources, constraining model size and capabilities, thus requiring optimized efficiency and prioritized knowledge. To address these challenges, we introduce Mobile-MMLU, a large-scale benchmark dataset tailored for mobile intelligence. It consists of 16,186 questions across 80 mobile-related fields, designed to evaluate LLM performance in realistic mobile scenarios. A challenging subset, Mobile-MMLU-Pro, provides advanced evaluation similar in size to MMLU-Pro but significantly more difficult than our standard full set. Both benchmarks use multiple-choice, order-invariant questions focused on practical mobile interactions, such as recipe suggestions, travel planning, and essential daily tasks. The dataset emphasizes critical mobile-specific metrics like inference latency, energy consumption, memory usage, and response quality, offering comprehensive insights into model performance under mobile constraints. Moreover, it prioritizes privacy and adaptability, assessing models' ability to perform on-device processing, maintain user privacy, and adapt to personalized usage patterns. Mobile-MMLU family offers a standardized framework for developing and comparing mobile-optimized LLMs, enabling advancements in productivity and decision-making within mobile computing environments. Our code and data are available at: https://github.com/VILA-Lab/Mobile-MMLU.