VEDA: Efficient LLM Generation Through Voting-based KV Cache Eviction and Dataflow-flexible Accelerator

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Wang, Zhican, Fan, Hongxiang, Waris, Haroon, Wang, Gang, Li, Zhenyu, Jiang, Jianfei, Sun, Yanan, He, Guanghui
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909670076055552
author Wang, Zhican
Fan, Hongxiang
Waris, Haroon
Wang, Gang
Li, Zhenyu
Jiang, Jianfei
Sun, Yanan
He, Guanghui
author_facet Wang, Zhican
Fan, Hongxiang
Waris, Haroon
Wang, Gang
Li, Zhenyu
Jiang, Jianfei
Sun, Yanan
He, Guanghui
contents Large Language Models (LLMs) excel in natural language processing tasks but pose significant computational and memory challenges for edge deployment due to their intensive resource demands. This work addresses the efficiency of LLM inference by algorithm-hardware-dataflow tri-optimizations. We propose a novel voting-based KV cache eviction algorithm, balancing hardware efficiency and algorithm accuracy by adaptively identifying unimportant kv vectors. From a dataflow perspective, we introduce a flexible-product dataflow and a runtime reconfigurable PE array for matrix-vector multiplication. The proposed approach effectively handles the diverse dimensional requirements and solves the challenges of incrementally varying sequence lengths. Additionally, an element-serial scheduling scheme is proposed for nonlinear operations, such as softmax and layer normalization (layernorm). Results demonstrate a substantial reduction in latency, accompanied by a significant decrease in hardware complexity, from O(N) to O(1). The proposed solution is realized in a custom-designed accelerator, VEDA, which outperforms existing hardware platforms. This research represents a significant advancement in LLM inference on resource-constrained edge devices, facilitating real-time processing, enhancing data privacy, and enabling model customization.
format Preprint
id arxiv_https___arxiv_org_abs_2507_00797
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle VEDA: Efficient LLM Generation Through Voting-based KV Cache Eviction and Dataflow-flexible Accelerator
Wang, Zhican
Fan, Hongxiang
Waris, Haroon
Wang, Gang
Li, Zhenyu
Jiang, Jianfei
Sun, Yanan
He, Guanghui
Hardware Architecture
Large Language Models (LLMs) excel in natural language processing tasks but pose significant computational and memory challenges for edge deployment due to their intensive resource demands. This work addresses the efficiency of LLM inference by algorithm-hardware-dataflow tri-optimizations. We propose a novel voting-based KV cache eviction algorithm, balancing hardware efficiency and algorithm accuracy by adaptively identifying unimportant kv vectors. From a dataflow perspective, we introduce a flexible-product dataflow and a runtime reconfigurable PE array for matrix-vector multiplication. The proposed approach effectively handles the diverse dimensional requirements and solves the challenges of incrementally varying sequence lengths. Additionally, an element-serial scheduling scheme is proposed for nonlinear operations, such as softmax and layer normalization (layernorm). Results demonstrate a substantial reduction in latency, accompanied by a significant decrease in hardware complexity, from O(N) to O(1). The proposed solution is realized in a custom-designed accelerator, VEDA, which outperforms existing hardware platforms. This research represents a significant advancement in LLM inference on resource-constrained edge devices, facilitating real-time processing, enhancing data privacy, and enabling model customization.
title VEDA: Efficient LLM Generation Through Voting-based KV Cache Eviction and Dataflow-flexible Accelerator
topic Hardware Architecture
url https://arxiv.org/abs/2507.00797