paper-with-me

Papers

LoongServe: Efficiently Serving Long-Context Large Language Models with Elastic Sequence Parallelism

2024-04-15 · Bingyang Wu, Shengyu Liu, Yinmin Zhong, Peng Sun, Xuanzhe Liu, Xin Jin

The context window of large language models (LLMs) is rapidly increasing, leading to a huge variance in resource usage between different requests as well as between different phases of the same request. Restricted by static parallelism strategies, existing LLM serving systems cannot efficiently utilize the underlying resources to serve variable-length requests in different phases. To address this problem, we propose a new parallelism paradigm, elastic sequence parallelism (ESP), to elastically adapt to the variance between different requests and phases. Based on ESP, we design and build LoongServe, an LLM serving system that (1) improves computation efficiency by elastically adjusting the degree of parallelism in real-time, (2) improves communication efficiency by reducing key-value cache migration overhead and overlapping partial decoding communication with computation, and (3) improves GPU memory efficiency by reducing key-value cache fragmentation across instances. Our evaluation under diverse real-world datasets shows that LoongServe improves the maximum throughput by up to 3.85$\times$ compared to the chunked prefill and 5.81$\times$ compared to the prefill-decoding disaggregation.

📄 PDF Abstract BibTeX arXiv:2404.09526

Code (1)

LoongServe/LoongServe 공식 구현 pytorch

Tasks

GPU

Methods 이 논문이 사용한 방법론

Hierarchical Feature Fusion Hierarchical Feature Fusion (HFF) is a feature fusion method employed in ESP and EESP image…
Dilated Convolution 설명 없음
Pointwise Convolution Pointwise Convolution is a type of convolution that uses a 1x1 kernel: a kernel that iterates through every single point. This…
Fragmentation Given a pattern $P,$ that is more complicated than the patterns, we fragment $P$ into simpler patterns such that their exact count is known. In the subgraph GNN proposed earlier,…
ESP 설명 없음

Similar Papers 제목 키워드 기반

LServe: Efficient Long-sequence LLM Serving with Unified Sparse Attention

2025-02-20 · Shang Yang, Junxian Guo, Haotian Tang, Qinghao Hu 외

Large language models (LLMs) have shown remarkable potential in processing long sequences and complex reasoning tasks, yet efficiently serving these models remains challenging due to the quadratic computational complexit…

Medha: Efficiently Serving Multi-Million Context Length LLM Inference Requests Without Approximations

2024-09-25 · Amey Agrawal, Haoran Qiu, Junda Chen, Íñigo Goiri 외

As large language models (LLMs) handle increasingly longer contexts, serving long inference requests of millions of tokens presents unique challenges. We show that existing work for long context inference is largely base…

BlockingChunkingScheduling

ParallelComp: Parallel Long-Context Compressor for Length Extrapolation

2025-02-20 · Jing Xiong, Jianghan Shen, Chuanyang Zheng, Zhongwei Wan 외

Efficiently handling long contexts is crucial for large language models (LLMs). While rotary position embeddings (RoPEs) enhance length generalization, effective length extrapolation remains challenging and often require…

4k8kGPU

Scaling Instruction-Tuned LLMs to Million-Token Contexts via Hierarchical Synthetic Data Generation

2025-04-17 · Linda He, Jue Wang, Maurice Weber, Shang Zhu 외

Large Language Models (LLMs) struggle with long-context reasoning, not only due to the quadratic scaling of computational complexity with sequence length but also because of the scarcity and expense of annotating long-co…

Synthetic Data Generation

FlexiCache: Leveraging Temporal Stability of Attention Heads for Efficient KV Cache Management

2025-11-02 · Nazmul Takbir, Hamidreza Alikhani, Nikil Dutt, Sangeetha Abdu Jyothi arxiv

Large Language Model (LLM) serving is increasingly constrained by the growing size of the key-value (KV) cache, which scales with both context length and generation length. Prior work shows that attention is dominated by…