# Splitwise: Collaborative Edge-Cloud Inference for LLMs via Lyapunov-Assisted DRL

arXiv: https://arxiv.org/abs/2512.23310 · HTML: https://mlsanity.com/paper/2512.23310/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 13 of its 35 references that are in the map.
- First posted on arXiv 29 Dec 2025.

## Builds on (most cited first, 13 of 13)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://mlsanity.com/paper/1704.04861/) (2017)
- [World Models](https://mlsanity.com/paper/1803.10122/) (2018)
- [LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Dataset](https://mlsanity.com/paper/2309.11998/) (2023)
- [Edge AI: On-Demand Accelerating Deep Neural Network Inference via Edge Computing](https://mlsanity.com/paper/1910.05316/) (2019)
- [SPINN: Synergistic Progressive Inference of Neural Networks over Device and Cloud](https://mlsanity.com/paper/2008.06402/) (2020)
- [CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge](https://mlsanity.com/paper/2506.02847/) (2025)
- [Jupiter: Fast and Resource-Efficient Collaborative Inference of Generative LLMs on Edge Devices](https://mlsanity.com/paper/2504.08242/) (2025)
- [PipeInfer: Accelerating LLM Inference using Asynchronous Pipelined Speculation](https://mlsanity.com/paper/2407.11798/) (2024)
- [Decentralized Low-Latency Collaborative Inference via Ensembles on the Edge](https://mlsanity.com/paper/2206.03165/) (2022)
- [Adaptive Resource Allocation Optimization Using Large Language Models in Dynamic Wireless Environments](https://mlsanity.com/paper/2502.02287/) (2025)
- [Large Language Model Partitioning for Low-Latency Inference at the Edge](https://mlsanity.com/paper/2505.02533/) (2025)
- [A Reinforcement Learning Formulation of the Lyapunov Optimization: Application to Edge Computing Systems with Queue Stability](https://mlsanity.com/paper/2012.07279/) (2020)

## Built on by (most cited first, 3 of 3)

- [Parameter- and Bandwidth-Efficient Edge--cloud Many-to-Many Speech-to-Text Translation](https://mlsanity.com/paper/2605.28642/) (2026)
- [MAPS: Memory-Aware Predictive Scheduling Framework for Large Language Model Serving](https://mlsanity.com/paper/2609.15359/) (2026)
- [End-to-End Latency-Minimizing and Load-Balanced Request Scheduling for Edge LLM Inference in Agentic AI Services](https://mlsanity.com/paper/2609.17193/) (2026)

## Abstract (arXiv)

Deploying large language models (LLMs) on edge devices is challenging due to their limited memory and power resources. Cloud-only inference reduces device burden but introduces high latency and cost. Static edge-cloud partitions optimize a single metric and struggle when bandwidth fluctuates. We propose Splitwise, a novel Lyapunov-assisted deep reinforcement learning (DRL) framework for fine-grained, adaptive partitioning of LLMs across edge and cloud environments. Splitwise decomposes transformer layers into attention heads and feed-forward sub-blocks, exposing more partition choices than layer-wise schemes. A hierarchical DRL policy, guided by Lyapunov optimization, jointly minimizes latency, energy consumption, and accuracy degradation while guaranteeing queue stability under stochastic workloads and variable network bandwidth. Splitwise also guarantees robustness via partition checkpoints with exponential backoff recovery in case of communication failures. Experiments on Jetson Orin NX, Galaxy S23, and Raspberry Pi 5 with GPT-2 (1.5B), LLaMA-7B, and LLaMA-13B show that Splitwise reduces end-to-end latency by 1.4x-2.8x and cuts energy consumption by up to 41% compared with existing partitioners. It lowers the 95th-percentile latency by 53-61% relative to cloud-only execution, while maintaining accuracy and modest memory requirements.