"A Dataset for Comparative Performance Analysis of On-Device and Edge-Assisted Small Language Model Inference on Android Devices"

Abstract

" This dataset presents experimental measurements for comparing on-device and edge-assisted text summarization using a small language model (SLM) on an Android device. The Android application is developed in Kotlin, with local inference performed using the MediaPipe LLM Inference runtime and edge inference supported by a remote server-based summarization service. The dataset contains 10,023 records, comprising 5,010 local-execution records and 5,013 edge-execution records, with 24 columns describing task characteristics, execution performance, device resources, battery conditions, and wireless network parameters. The recorded attributes include input word count, execution latency, CPU utilization, memory consumption, battery level, battery temperature, Wi-Fi received signal strength indicator (RSSI), Wi-Fi round-trip time (RTT), and upload and download throughput. The experiments use the Llama-3.2-1B-Instruct-Q8_0 model to investigate the performance characteristics of local and edge-based summarization under the recorded experimental conditions. The dataset supports comparative latency analysis, investigation of relationships between device and network conditions and execution performance, and the development of machine-learning models for adaptive computation offloading. It provides a resource for research on mobile artificial intelligence, edge intelligence, and resource-aware task placement, particularly for determining when language-model tasks should be processed locally or offloaded to an edge server."