A Dataset for Computation Offloading Cost of Large Language Model Inference in Edge Computing

OverviewThis dataset focuses on the computation offloading problem of large model inference in edge computing scenarios, providing standardized data support for related algorithm development and performance evaluation.Main Contributions of the DatasetAddresses the computationally intensive and latency-sensitive characteristics of large model inference tasks, focusing on computational offloading requirements in resource-constrained edge computing.Fills the gap in existing datasets for the interdisciplinary scenario of "large model inference + edge computing + offloading decision".

Supporting research on offloading strategy optimization and resource scheduling algorithms.Data ContentHardware resources and energy consumption data, including utilization parameters of GPU devices such as RTX 4090 and P2000, as well as total energy consumption throughout the inference process and phased energy consumption data (loading/prompt/generation).Data on large model inference tasks, covering model scales (gemma3, qwen3, llama3.2, etc.), inference latency (total task time, model processing time, etc.), and data transmission volume (prompt/generation/total tokens count) and other key metrics.Quality and performance feedback data for large model inference, such as inference result correctness (correct/incorrect), end-to-end timing nodes (task initiation/processing/return timestamps), and resource consumption efficiency (token processing speed).Application of the DatasetEnables verification of computational offloading algorithm effectiveness and comparison of different offloading strategies in terms of latency, energy consumption, and resource utilization.Provides data benchmarks for research on resource scheduling and offloading decision optimization for large model inference in edge computing environments, facilitating the implementation of related technologies.Additional