paper-with-me

홈 › Papers

Compressing DMA Engine: Leveraging Activation Sparsity for Training Deep Neural Networks

2017-05-03 · Minsoo Rhu, Mike O'Connor, Niladrish Chatterjee, Jeff Pool, Stephen W. Keckler

Popular deep learning frameworks require users to fine-tune their memory usage so that the training data of a deep neural network (DNN) fits within the GPU physical memory. Prior work tries to address this restriction by virtualizing the memory usage of DNNs, enabling both CPU and GPU memory to be utilized for memory allocations. Despite its merits, virtualizing memory can incur significant performance overheads when the time needed to copy data back and forth from CPU memory is higher than the latency to perform the computations required for DNN forward and backward propagation. We introduce a high-performance virtualization strategy based on a "compressing DMA engine" (cDMA) that drastically reduces the size of the data structures that are targeted for CPU-side allocations. The cDMA engine offers an average 2.6x (maximum 13.8x) compression ratio by exploiting the sparsity inherent in offloaded data, improving the performance of virtualized DNNs by an average 32% (maximum 61%).

📄 PDF Abstract BibTeX arXiv:1705.01626

Code (0)

등록된 구현이 없습니다.

Tasks

CPUGPU

Similar Papers 제목 키워드 기반

SoLA: Leveraging Soft Activation Sparsity and Low-Rank Decomposition for Large Language Model Compression

2026-03-12 · Xinhao Huang, You-Liang Huang, Zeyi Wen arxiv

Large language models (LLMs) have demonstrated impressive capabilities across various tasks, but the billion-scale parameters pose deployment challenges. Although existing methods attempt to reduce the scale of LLMs, the…

Model Compression

Compressing the Backward Pass of Large-Scale Neural Architectures by Structured Activation Pruning

2023-11-28 · Daniel Barley, Holger Fröning

The rise of Deep Neural Networks (DNNs) has led to an increase in model size and complexity, straining the memory capacity of GPUs. Sparsity in DNNs, characterized as structural or ephemeral, has gained attention as a so…

GPUimage-classificationImage Classification

COAT: Compressing Optimizer states and Activation for Memory-Efficient FP8 Training

2024-10-25 · Haocheng Xi, Han Cai, Ligeng Zhu, Yao Lu 외

FP8 training has emerged as a promising method for improving training efficiency. Existing frameworks accelerate training by applying FP8 computation to linear layers while leaving optimizer states and activations in hig…

Language ModelingLanguage ModellingLarge Language ModelQuantization

Amber Pruner: Leveraging N:M Activation Sparsity for Efficient Prefill in Large Language Models

2025-08-04 · Tai An, Ruwu Cai, Yanzhe Zhang, Yang Liu 외 arxiv

In the era of large language models (LLMs), N:M sparsity has emerged as a structured compression technique critical for accelerating inference. While prior work has primarily focused on weight sparsity, it often suffers …

Activation Sparsity Opportunities for Compressing General Large Language Models

2024-12-13 · Nobel Dhar, Bobin Deng, Md Romyull Islam, Kazi Fahim Ahmad Nasif 외

Deploying local AI models, such as Large Language Models (LLMs), to edge devices can substantially enhance devices' independent capabilities, alleviate the server's burden, and lower the response time. Owing to these tre…

Model Compression