Cover Image

DynaKV: Enabling Accurate and Efficient Long-Sequence LLM Decoding on Smartphones

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Wang, Tuowei, Huang, Minxing, Li, Fengzu, Chen, Ligeng, Zhang, Jinrui, Ren, Ju
Format:	Preprint
Published:	2025
Subjects:	Distributed, Parallel, and Cluster Computing Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Scaling LLM Test-Time Compute with Mobile NPU on Smartphones
by: Hao, Zixu, et al.
Published: (2025)

MoSKA: Mixture of Shared KV Attention for Efficient Long-Sequence LLM Inference
by: Rhee, Myunghyun, et al.
Published: (2025)

FlowSpec: Continuous Pipelined Speculative Decoding for Efficient Distributed LLM Inference
by: Liu, Xing, et al.
Published: (2025)

SnapStream: Efficient Long Sequence Decoding on Dataflow Accelerators
by: Li, Jonathan, et al.
Published: (2025)

PackKV: Reducing KV Cache Memory Footprint through LLM-Aware Lossy Compression
by: Jiang, Bo, et al.
Published: (2025)

KVSwap: Disk-aware KV Cache Offloading for Long-Context On-device Inference
by: Zhang, Huawei, et al.
Published: (2025)

KVComp: A High-Performance, LLM-Aware, Lossy Compression Framework for KV Cache
by: Jiang, Bo, et al.
Published: (2025)

FairKV: Balancing Per-Head KV Cache for Fast Multi-GPU Inference
by: Zhao, Bingzhe, et al.
Published: (2025)

Speculative Decoding in Decentralized LLM Inference: Turning Communication Latency into Computation Throughput
by: Song, Jingwei, et al.
Published: (2025)

InfiniPipe: Elastic Pipeline Parallelism for Efficient Variable-Length Long-Context LLM Training
by: Wang, Shiju, et al.
Published: (2025)

PipeSpec: Breaking Stage Dependencies in Hierarchical LLM Decoding
by: McDanel, Bradley, et al.
Published: (2025)

A Survey on Large Language Model Acceleration based on KV Cache Management
by: Li, Haoyang, et al.
Published: (2024)

Helix Parallelism: Rethinking Sharding Strategies for Interactive Multi-Million-Token LLM Decoding
by: Bhatia, Nidhi, et al.
Published: (2025)

DUAL-BLADE: Dual-Path NVMe-Direct KV-Cache Offloading for Edge LLM Inference
by: Jeong, Bodon, et al.
Published: (2026)

KVServe: Service-Aware KV Cache Compression for Communication-Efficient Disaggregated LLM Serving
by: Liu, Zedong, et al.
Published: (2026)

Striking the Right Balance between Compute and Copy: Improving LLM Inferencing Under Speculative Decoding
by: Ramachandran, Arun, et al.
Published: (2025)

SPECTRE: Hybrid Ordinary-Parallel Speculative Serving for Resource-Efficient LLM Inference
by: Xie, Jincheng, et al.
Published: (2026)

Token-Budget-Aware Pool Routing for Cost-Efficient LLM Inference
by: Chen, Huamin, et al.
Published: (2026)

Efficient Multi-Adapter LLM Serving via Cross-Model KV-Cache Reuse with Activated LoRA
by: Li, Allison, et al.
Published: (2025)

MatKV: Trading Compute for Flash Storage in LLM Inference
by: Shin, Kun-Woo, et al.
Published: (2025)

KunServe: Parameter-centric Memory Management for Efficient Memory Overloading Handling in LLM Serving
by: Cheng, Rongxin, et al.
Published: (2024)

TACO: Efficient Communication Compression of Intermediate Tensors for Scalable Tensor-Parallel LLM Training
by: Liu, Man, et al.
Published: (2026)

PRESERVE: Prefetching Model Weights and KV-Cache in Distributed LLM Serving
by: Yüzügüler, Ahmet Caner, et al.
Published: (2025)

High-Throughput LLM inference on Heterogeneous Clusters
by: Xiong, Yi, et al.
Published: (2025)

SplitZip: Ultra Fast Lossless KV Compression for Disaggregated LLM Serving
by: Guo, Yipin, et al.
Published: (2026)

PiKV: KV Cache Management System for Mixture of Experts
by: Liu, Dong, et al.
Published: (2025)

ENOVA: Autoscaling towards Cost-effective and Stable Serverless LLM Serving
by: Huang, Tao, et al.
Published: (2024)

TinyServe: Query-Aware Cache Selection for Efficient LLM Serving
by: Liu, Dong, et al.
Published: (2025)

Lumos: Efficient Performance Modeling and Estimation for Large-scale LLM Training
by: Liang, Mingyu, et al.
Published: (2025)

KV-Runahead: Scalable Causal LLM Inference by Parallel Key-Value Cache Generation
by: Cho, Minsik, et al.
Published: (2024)

Beluga: A CXL-Based Memory Architecture for Scalable and Efficient LLM KVCache Management
by: Yang, Xinjun, et al.
Published: (2025)

BucketServe: Bucket-Based Dynamic Batching for Smart and Efficient LLM Inference Serving
by: Zheng, Wanyi, et al.
Published: (2025)

Elastic On-Device LLM Service
by: Yin, Wangsong, et al.
Published: (2024)

KAITIAN: A Unified Communication Framework for Enabling Efficient Collaboration Across Heterogeneous Accelerators in Embodied AI Systems
by: Lin, Jieke, et al.
Published: (2025)

A Few GPUs, A Whole Lotta Scale: Faithful LLM Training Emulation with PrismLLM
by: Xi, Shaoke, et al.
Published: (2026)

Accurate GPU Memory Prediction for Deep Learning Jobs through Dynamic Analysis
by: Shi, Jiabo, et al.
Published: (2025)

Frontier: Towards Comprehensive and Accurate LLM Inference Simulation
by: Feng, Yicheng, et al.
Published: (2026)

xLLM Technical Report
by: Liu, Tongxuan, et al.
Published: (2025)

Revisiting Parameter Server in LLM Post-Training
by: Wan, Xinyi, et al.
Published: (2026)

SPEED-Bench: A Unified and Diverse Benchmark for Speculative Decoding
by: Abramovich, Talor, et al.
Published: (2026)