# Efficient Multi-Adapter LLM Serving via Cross-Model KV-Cache Reuse with Activated LoRA

arXiv: https://arxiv.org/abs/2512.17910 · HTML: https://mlsanity.com/paper/2512.17910/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Placed in Serving systems and request scheduling, one of 12 directions in Efficient LLM inference and serving.
- Builds on 13 of its 20 references that are in the map.
- First posted on arXiv 26 Nov 2025.

## 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, 13 of 13)

- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality](https://mlsanity.com/paper/2405.21060/) (2024)
- [LLM-Planner: Few-Shot Grounded Planning for Embodied Agents with Large Language Models](https://mlsanity.com/paper/2212.04088/) (2022)
- [SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills](https://mlsanity.com/paper/2308.16369/) (2023)
- [ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learning](https://mlsanity.com/paper/2503.19470/) (2025)
- [SWEET-RL: Training Multi-Turn LLM Agents on Collaborative Reasoning Tasks](https://mlsanity.com/paper/2503.15478/) (2025)
- [A Closer Look at the Limitations of Instruction Tuning](https://mlsanity.com/paper/2402.05119/) (2024)
- [Neeko: Leveraging Dynamic LoRA for Efficient Multi-Character Role-Playing Agent](https://mlsanity.com/paper/2402.13717/) (2024)
- [Activation-Informed Merging of Large Language Models](https://mlsanity.com/paper/2502.02421/) (2025)
- [CaraServe: CPU-Assisted and Rank-Aware LoRA Serving for Generative LLM Inference](https://mlsanity.com/paper/2401.11240/) (2024)
- [Activated LoRA: Fine-tuned LLMs for Intrinsics](https://mlsanity.com/paper/2504.12397/) (2025)
- [Instruct Once, Chat Consistently in Multiple Rounds: An Efficient Tuning Framework for Dialogue](https://mlsanity.com/paper/2402.06967/) (2024)
- [MALM: A Multi-Information Adapter for Large Language Models to Mitigate Hallucination](https://mlsanity.com/paper/2506.12483/) (2025)

## Built on by (most cited first, 8 of 8)

- [CacheBridge: Efficient Cross-Model KV Cache Transfer](https://mlsanity.com/paper/2609.00891/) (2026)
- [Rethinking LoRA Memory Through the Lens of KV Cache Compression](https://mlsanity.com/paper/2606.05698/) (2026)
- [From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves](https://mlsanity.com/paper/2602.24210/) (2026)
- [Agents unlock new capabilities through Switching LoRA Adapters as a Tool (SLAaaT)](https://mlsanity.com/paper/2608.17034/) (2026)
- [Unified AI Gateway: A Framework for Joint Model Routing and KV Cache Management](https://mlsanity.com/paper/2609.06940/) (2026)
- [An Efficient and Modular Framework for Targeted Harm Mitigation in LLMS](https://mlsanity.com/paper/2609.13624/) (2026)
- [Shared-Prefix KV Reuse Across Standard LoRA Adapters: Quality and Serving Tradeoffs](https://mlsanity.com/paper/2609.17109/) (2026)
- [To Cache or Not to Cache: Understanding and Mitigating Safety Incompatibility in Cross-Model KV Cache Reuse](https://mlsanity.com/paper/2610.15721/) (2026)

## Abstract (arXiv)

Modern large language model (LLM) systems increasingly rely on multi-turn pipelines that are composed of multiple task-specific adapters, yet existing serving frameworks remain inefficient, incurring substantial recomputation overhead when switching between adapters. We present the first LLM serving engine that supports cross-model prefix cache reuse between base and adapted models via Activated LoRA (aLoRA), enabling efficient and fine-grained adapter switching during inference. Our design extends the vLLM framework by introducing base-aligned block hashing and activation-aware masking within the model execution path, permitting cache reuse across models while preserving compatibility with existing serving engine optimizations. Integrated into a production-grade inference stack, this approach supports dynamic adapter activation without excessive key-value tensor recomputation. Evaluation across representative multi-turn, multi-adapter pipelines demonstrates up to 58x end-to-end latency reduction and over 100x time-to-first-token improvement relative to standard LoRA baselines, with benefits that scale with model size and sequence length and manifest across all stages of the request lifecycle. This work bridges parameter-efficient model adaptation with high-performance serving, providing the first complete realization of cross-model KV-cache reuse in modern LLM inference engines.