paper-with-me

Papers

PyramidInfer: Pyramid KV Cache Compression for High-throughput LLM Inference

2024-05-21 · Dongjie Yang, Xiaodong Han, Yan Gao, Yao Hu, Shilin Zhang, Hai Zhao

Large Language Models (LLMs) have shown remarkable comprehension abilities but face challenges in GPU memory usage during inference, hindering their scalability for real-time applications like chatbots. To accelerate inference, we store computed keys and values (KV cache) in the GPU memory. Existing methods study the KV cache compression to reduce memory by pruning the pre-computed KV cache. However, they neglect the inter-layer dependency between layers and huge memory consumption in pre-computation. To explore these deficiencies, we find that the number of crucial keys and values that influence future generations decreases layer by layer and we can extract them by the consistency in attention weights. Based on the findings, we propose PyramidInfer, a method that compresses the KV cache by layer-wise retaining crucial context. PyramidInfer saves significant memory by computing fewer keys and values without sacrificing performance. Experimental results show PyramidInfer improves 2.2x throughput compared to Accelerate with over 54% GPU memory reduction in KV cache.

📄 PDF Abstract BibTeX arXiv:2405.12532

Code (2)

mutonix/pyramidinfer pytorch
tyxqc/spindlekv pytorch

Tasks

GPU

Methods 이 논문이 사용한 방법론

Pruning 설명 없음

Similar Papers 제목 키워드 기반

LightVLM: Acceleraing Large Multimodal Models with Pyramid Token Merging and KV Cache Compression

2025-08-30 · Lianyu Hu, Fanhua Shang, Wei Feng, Liang Wan arxiv

In this paper, we introduce LightVLM, a simple but effective method that can be seamlessly deployed upon existing Vision-Language Models (VLMs) to greatly accelerate the inference process in a training-free manner. We di…

PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling

2024-06-04 · Zefan Cai, Yichi Zhang, Bofei Gao, Yuliang Liu 외

In this study, we investigate whether attention-based information flow inside large language models (LLMs) is aggregated through noticeable patterns for long context processing. Our observations reveal that LLMs aggregat…

KARA: Efficient Reasoning LLM Serving via Sliding-Window KV Cache Compression

2026-05-01 · Shen Han, Yuyang Wu, Junpu Yu, Olexandr Isayev arxiv

Reasoning language models often generate long chain-of-thought (CoT), which accumulates a massive KV cache during the decoding phase and incurs high decoding latency and limited throughput. To address these issues, KV ca…

Joint Encoding of KV-Cache Blocks for Scalable LLM Serving

2026-01-06 · Joseph Kampeas, Emir Haleva arxiv

Modern large language models (LLMs) drive interactive AI systems but are bottlenecked by the memory-heavy growth of key-value (KV) caches, which limits real-time throughput under concurrent loads. Existing KV-cache compr…

MosaicKV: Serving Long-Context LLM with Dynamic Two-D KV Cache Compression

2026-07-01 · Sheng Qiang, Ruiwei Chen, Yinpeng Wu, Jinyu Gu 외 arxiv

Long-context LLM services now sustain prompts with hundreds of thousands to millions of tokens, making the key-value (KV) cache a first-order serving cost. Because the cache grows linearly with context length, it can exh…