Mobile Edge Intelligence for LLM Inference: Offloading Mechanisms, Resource Optimization, and Open Challenges

This review examined how mobile edge computing can make large language model inference practical for mobile users. We showed that LLM inference differs from conventional DNN offloading because of its autoregressive decoding, stateful KV cache, and extreme model size [5], [13]. We organized offloading strategies into full offloading, split inference, collaborative inference, and hybrid draft–verify or routing schemes [8]–[11], [15]–[20], and reviewed resource optimization across model compression [21]–[24], memory management [12]–[14], [25], computation scheduling, communication, and energy. We also summarized enabling systems, applications, and evaluation metrics. The analysis suggests that no single technique is sufficient. Efficient edge LLM services will combine compressed models, memory-efficient serving, adaptive offloading, and network-aware scheduling within a joint optimization framework. Key open problems include token-level adaptive offloading, KV cache mobility, privacy of intermediate activations, hardware heterogeneity, MoE deployment, cross-layer 6G co-design, standardized benchmarking, and sustainability. Addressing these challenges will be central to realizing the vision of edge-native generative AI in future mobile networks [8], [9].

[1] Satyanarayanan, M. (2017). The emergence of edge computing. Computer, 50(1), 30–39. https://doi.org/10.1109/MC.2017.9 [2] Mao, Y., You, C., Zhang, J., Huang, K., & Letaief, K. B. (2017). A survey on mobile edge computing: The communication perspective. IEEE Communications Surveys & Tutorials, 19(4), 2322–2358. https://doi.org/10.1109/COMST.2017.2745201 [3] Shi, W., Cao, J., Zhang, Q., Li, Y., & Xu, L. (2016). Edge computing: Vision and challenges. IEEE Internet of Things Journal, 3(5), 637–646. https://doi.org/10.1109/JIOT.2016.2579198 [4] Mach, P., & Becvar, Z. (2017). Mobile edge computing: A survey on architecture and computation offloading. IEEE Communications Surveys & Tutorials, 19(3), 1628–1656. https://doi.org/10.1109/COMST.2017.2682318 [5] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. In Advances in Neural Information Processing Systems (Vol. 30). Curran Associates. [6] Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., Agarwal, S., Herbert-Voss, A., Krueger, G., Henighan, T., Child, R., Ramesh, A., Ziegler, D. M., Wu, J., Winter, C., . . . Amodei, D. (2020). Language models are few-shot learners. In Advances in Neural Information Processing Systems (Vol. 33, pp. 1877–1901). Curran Associates. [7] Zhou, Z., Chen, X., Li, E., Zeng, L., Luo, K., & Zhang, J. (2019). Edge intelligence: Paving the last mile of artificial intelligence with edge computing. Proceedings of the IEEE, 107(8), 1738–1762. https://doi.org/10.1109/JPROC.2019.2918951 [8] Lin, Z., Qu, G., Chen, Q., Chen, X., Chen, Z., & Huang, K. (2025). Pushing large language models to the 6G edge: Vision, challenges, and opportunities. IEEE Communications Magazine, 63(9), 52–59. [9] Xu, M., Du, H., Niyato, D., Kang, J., Xiong, Z., Mao, S., Han, Z., Jamalipour, A., Kim, D. I., Shen, X., Leung, V. C. M., & Poor, H. V. (2024). Unleashing the power of edge-cloud generative AI in mobile networks: A survey of AIGC services. IEEE Communications Surveys & Tutorials, 26(2), 1127–1170. https://doi.org/10.1109/COMST.2024.3353265 [10] Kang, Y., Hauswald, J., Gao, C., Rovinski, A., Mudge, T., Mars, J., & Tang, L. (2017). Neurosurgeon: Collaborative intelligence between the cloud and mobile edge. In Proceedings of the 22nd International Conference on Architectural Support for Programming Languages and Operating Systems (pp. 615–629). ACM. https://doi.org/10.1145/3037697.3037698