面向边缘设备的多 LoRA 基于统一自适应蒸馏的量化：实现基于 one-for-all 的生成式视觉模型
摘要
Generative Artificial Intelligence (GenAI) 如图像编辑、对象移除和 prompt-guided image transformation 日益集成到 mobile applications 中。然而，在 resource-constrained devices 上部署 Large Vision Models (LVMs) 仍然具有挑战性 due to their high memory and compute requirements。虽然 Low-Rank Adapters (LoRAs) enable parameter-efficient task adaptation，但 existing Mobile deployment pipelines typically compile separate model binaries for each LoRA + a copy of the foundation model， resulting in redundant storage and increased runtime overhead。In this work, we present a unified framework for enabling multi-task GenAI inference on edge devices using a single shared model。Our key idea is to treat LoRA weights as runtime inputs rather than embedding them into the compiled model graph，allowing dynamic task switching at runtime without recompilation。Then, to support efficient on-device execution，we introduce QUAD (Quantization with Unified Adaptive Distillation)，a quantization-aware training strategy that aligns multiple LoRA adapters under a shared quantization profile。We implement the proposed system with a lightweight runtime stack compatible with mobile NPUs and evaluate it across multiple chipsets。Experimental results demonstrate up to 6x and 4x reduction in memory footprint and latency improvements，respectively，while maintaining high visual quality across multiple GenAI tasks。
引用
@article{arxiv.2603.29535,
title = {Quantization with Unified Adaptive Distillation to enable multi-LoRA based one-for-all Generative Vision Models on edge},
author = {Sowmya Vajrala and Aakash Parmar and Prasanna R and Sravanth Kodavanti and Manjunath Arveti and Srinivas Soumitri Miriyala and Ashok Senapati},
journal= {arXiv preprint arXiv:2603.29535},
year = {2026}
}
备注
Accepted at the Mobile AI Workshop, CVPR 2026