# POLAR: Online Learning for LoRA Adapter Caching and Routing in Edge LLM Serving

arXiv: https://arxiv.org/abs/2604.16583 · HTML: https://mlsanity.com/paper/2604.16583/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Serving systems and request scheduling, one of 12 directions in Efficient LLM inference and serving.
- Builds on 11 of its 27 references that are in the map.
- First posted on arXiv 17 Apr 2026.

## Placed in

- Efficient LLM inference and serving / [Serving systems and request scheduling](https://mlsanity.com/inference/serving-systems-and-request-scheduling/)

## Builds on (most cited first, 11 of 11)

- [S-LoRA: Serving Thousands of Concurrent LoRA Adapters](https://mlsanity.com/paper/2311.03285/) (2023)
- [LLM Bandit: Cost-Efficient LLM Generation via Preference-Conditioned Dynamic Routing](https://mlsanity.com/paper/2502.02743/) (2025)
- [CaraServe: CPU-Assisted and Rank-Aware LoRA Serving for Generative LLM Inference](https://mlsanity.com/paper/2401.11240/) (2024)
- [Online Multi-LLM Selection via Contextual Bandits under Unstructured Context Evolution](https://mlsanity.com/paper/2506.17670/) (2025)
- [CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge](https://mlsanity.com/paper/2506.02847/) (2025)
- [Learning to Route LLMs from Bandit Feedback: One Policy, Many Trade-offs](https://mlsanity.com/paper/2510.07429/) (2025)
- [Semantic Caching for Low-Cost LLM Serving: From Offline Learning to Online Adaptation](https://mlsanity.com/paper/2508.07675/) (2025)
- [EdgeLoRA: An Efficient Multi-Tenant LLM Serving System on Edge Devices](https://mlsanity.com/paper/2507.01438/) (2025)
- [Serving Long-Context LLMs at the Mobile Edge: Test-Time Reinforcement Learning-based Model Caching and Inference Offloading](https://mlsanity.com/paper/2501.14205/) (2025)
- [Near-Optimal Online Deployment and Routing for Streaming LLMs](https://mlsanity.com/paper/2506.17254/) (2025)
- [Adapter-Augmented Bandits for Online Multi-Constrained Multi-Modal Inference Scheduling](https://mlsanity.com/paper/2603.06403/) (2026)

## Abstract (arXiv)

Edge deployment of large language models (LLMs) increasingly relies on libraries of lightweight LoRA adapters, yet GPU/DRAM can keep only a small resident subset at a time. Serving a request through a non-resident adapter requires paging its weights from storage, incurring measurable latency. This creates a two-timescale online control problem: on a slow timescale, the system selects which adapters remain resident in fast memory, while on a fast timescale it routes each request to an adapter whose context-dependent utility is unknown a priori. The two decisions are tightly coupled: the cache determines the cost of exploration, and the router determines which adapters receive informative feedback. We formulate this joint caching-and-routing problem as a two-timescale contextual bandit and propose POLAR (Paging and Online Learning for Adapter Routing). POLAR pairs a cache-aware LinUCB router with an epoch-based cache controller. We study two variants. A fixed-epoch version provides a robust baseline with worst-case regret guarantees under arbitrary contexts. An epoch-doubling version, POLAR+, adds forced exploration and improved cache optimization to achieve widetildeO(d√NT+√KT) sublinear regret under stochastic regularity and cacheability conditions, where N is the adapter count, K the cache size, d the context dimension, and T the horizon. The routing term matches the standard contextual-bandit rate up to logarithmic factors, showing that the memory hierarchy does not fundamentally slow routing learning. Experiments using 15 real LoRA adapters for Qwen2.5-7B together with measured GPU paging latencies show that adaptive cache control substantially outperforms non-adaptive baselines and exhibits scaling trends consistent with the theory.