mzCache: On-Device LLM Memory Management under Multitasking

• URL: http://arxiv.org/abs/2609.01338v1
• Date: Tue, 01 Sep 2026 14:49:20 GMT
• Title: mzCache: On-Device LLM Memory Management under Multitasking
• Abstract summary: We present mzCache, an on-device mobile Large Language Model (LLM) inference system with specialized memory management.<n>Under unpredictable memory pressure, mzCache elastically evicts LLM memory and the unified memory of mobile SOCs.<n>mzCache achieves 2.1-5.5$$ reduction in Time-to-First-Token compared to storage-backed partial offload.
• Score: 6.974241731162878
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: On-device mobile Large Language Model (LLM) inference is gaining significant attention. However, mobile devices operate in highly dynamic multitasking environments where users frequently switch between applications. This creates memory pressure, forcing LLM memory (model weights and KV cache) to be evicted by the operating system. When a new inference request arrives, the inference system must restore the evicted memory through slow storage reads or recompute the entire KV cache, severely degrading responsiveness. To address this, we present mzCache, an on-device LLM inference system with specialized memory management for multitasking environments. Under unpredictable memory pressure, mzCache elastically evicts LLM memory and leverages the unified memory of mobile SoCs to enable zero-wait inference on the GPU with concurrent CPU-side restoration. mzCache realizes this through restoration-oriented memory management: LLM memory is partitioned into fine-grained shared buffers to enable partial eviction and restoration with concurrent cross-processor access, while hybrid swap and backward-out eviction policies ensure low-latency restoration from any eviction state. Implemented on llama.cpp and deployed as an Android application, mzCache achieves 2.1-5.5$\times$ reduction in Time-to-First-Token compared to storage-backed partial offload and demonstrates its effectiveness in real multitasking scenarios.

Related papers

• Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents [72.3752981234916]
  Large Language Model (LLM) agents increasingly rely on external memory systems to accumulate experience across tasks.<n>We present Memory as a Controlled Process (MemCon), a framework that models memory operations as a Markov Decision Process.<n>MemCon consistently outperforms multiple memory baselines by up to 15.2 points in task success while reducing token consumption by 5-20%.
  arXiv  Detail & Related papers  (2026-07-15T08:32:40Z)
• CALMem : Application-Layer Dual Memory for Conversational AI [0.0]
  CALMem is an application-layer dual memory architecture that gives conversational assistants virtually effective context without any modification to the underlying model.<n>A token-budget-adaptive injection mechanism, called the MOIM, automatically retrieves and injects relevant past context each turn.<n>We describe the architecture, design decisions, and performance characteristics, and analyse the trade-offs that guided each implementation choice.
  arXiv  Detail & Related papers  (2026-05-20T05:23:57Z)
• RecMem: Recurrence-based Memory Consolidation for Efficient and Effective Long-Running LLM Agents [16.625864201773343]
  RecMem stores incoming interactions in a subconscious memory layer and encodes them using lightweight embedding models for retrieval.<n>Experiments show that RecMem reduces the memory construction token cost of three SOTA memory systems by up to 87% while exceeding their accuracy.
  arXiv  Detail & Related papers  (2026-05-15T15:17:36Z)
• MemMA: Coordinating the Memory Cycle through Multi-Agent Reasoning and In-Situ Self-Evolution [52.29303869050117]
  Memory-augmented LLM agents maintain external memory banks to support long-horizon interaction.<n>MemMA is a plug-and-play multi-agent framework that coordinates the memory cycle along both the forward and backward paths.
  arXiv  Detail & Related papers  (2026-03-19T10:15:59Z)
• MemLoRA: Distilling Expert Adapters for On-Device Memory Systems [71.32550994522738]
  Memory-augmented Large Language Models (LLMs) have demonstrated remarkable consistency during dialogues.<n>MemLoRA is a novel memory system that integrates small Vision-Language Models.<n>VLM-integrated MemLoRA-V shows massive improvements in caption-based approaches.
  arXiv  Detail & Related papers  (2025-12-04T12:56:30Z)
• KVSwap: Disk-aware KV Cache Offloading for Long-Context On-device Inference [6.159622195480178]
  Language models (LMs) underpin emerging mobile and embedded AI applications like meeting and video summarization and document analysis.<n>Long-context inference quickly hits a emphmemory capacity wall as the key-value ( KV) cache grows linearly with context length and batch size.<n>We present KVSwap, a software framework to break this memory wall by offloading the KV cache to non-volatile secondary storage (disk)<n> KVSwap delivers higher throughput under tight memory budgets while maintaining the generation quality when compared with existing KV cache offloading schemes.
  arXiv  Detail & Related papers  (2025-11-14T22:37:57Z)
• Sparse-dLLM: Accelerating Diffusion LLMs with Dynamic Cache Eviction [72.27673320976933]
  Diffusion Large Language Models (dLLMs) enable breakthroughs in reasoning and parallel decoding.<n>Current caching techniques accelerate decoding by storing full-layer states, yet impose substantial memory usage.<n>We propose Sparse-dLLM, the first training-free framework integrating dynamic cache eviction with sparse attention.
  arXiv  Detail & Related papers  (2025-08-04T16:14:03Z)
• vTensor: Flexible Virtual Tensor Management for Efficient LLM Serving [53.972175896814505]
  Large Language Models (LLMs) are widely used across various domains, processing millions of daily requests. Large Language Models (LLMs) are widely used across various domains, processing millions of daily requests.
  arXiv  Detail & Related papers  (2024-07-22T14:37:58Z)
• vAttention: Dynamic Memory Management for Serving LLMs without PagedAttention [8.20523619534105]
  PagedAttention is a popular approach for dynamic memory allocation in LLM serving systems.<n>We present vAttention -- an approach that mitigates fragmentation in physical memory while retaining the contiguity of KV cache in virtual memory.<n>Overall, vAttention is a simpler, portable, and performant alternative to PagedAttention.
  arXiv  Detail & Related papers  (2024-05-07T16:00:32Z)
• Efficient Memory Management for Large Language Model Serving with PagedAttention [44.70922552274376]
  High throughput serving of large language models (LLMs) requires sufficiently many requests at a time. Existing systems struggle because the key-value cache ( KV cache) memory for each request is huge and grows and shrinks dynamically. We propose PagedAttention, an attention algorithm inspired by the classical virtual memory and paging techniques in operating systems.
  arXiv  Detail & Related papers  (2023-09-12T12:50:04Z)
• Recurrent Dynamic Embedding for Video Object Segmentation [54.52527157232795]
  We propose a Recurrent Dynamic Embedding (RDE) to build a memory bank of constant size. We propose an unbiased guidance loss during the training stage, which makes SAM more robust in long videos. We also design a novel self-correction strategy so that the network can repair the embeddings of masks with different qualities in the memory bank.
  arXiv  Detail & Related papers  (2022-05-08T02:24:43Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.