paper-with-me

Papers

A Graph Theoretic Framework of Recomputation Algorithms for Memory-Efficient Backpropagation

2019-05-28 · NeurIPS 2019 12 · Mitsuru Kusumoto, Takuya Inoue, Gentaro Watanabe, Takuya Akiba, Masanori Koyama

Recomputation algorithms collectively refer to a family of methods that aims to reduce the memory consumption of the backpropagation by selectively discarding the intermediate results of the forward propagation and recomputing the discarded results as needed. In this paper, we will propose a novel and efficient recomputation method that can be applied to a wider range of neural nets than previous methods. We use the language of graph theory to formalize the general recomputation problem of minimizing the computational overhead under a fixed memory budget constraint, and provide a dynamic programming solution to the problem. Our method can reduce the peak memory consumption on various benchmark networks by 36%~81%, which outperforms the reduction achieved by other methods.

📄 PDF Abstract BibTeX arXiv:1905.11722

Code (3)

PaddlePaddle/FleetX/blob/develop/examples/resnet/train_fleet_recompute.py paddle
PaddlePaddle/FleetX/tree/develop/examples/recompute paddle
pfnet-research/recompute

Similar Papers 제목 키워드 기반

Adacc: An Adaptive Framework Unifying Compression and Activation Recomputation for LLM Training

2025-08-01 · Ping Chen, Zhuohong Deng, Ping Li, Shuibing He 외 arxiv

Training large language models (LLMs) is often constrained by GPU memory limitations. To alleviate memory pressure, activation recomputation and data compression have been proposed as two major strategies. However, both …

Optimizing Large Model Training through Overlapped Activation Recomputation

2024-06-13 · Ping Chen, Wenjie Zhang, Shuibing He, Weijian Chen 외

Large model training often uses recomputation to alleviate memory pressure and pipelines to exploit the parallelism of data, tensors, and devices. However, existing recomputation approaches may incur high overhead when t…

modelScheduling

DELTA: Dynamically Optimizing GPU Memory beyond Tensor Recomputation

2022-03-30 · Yu Tang, Chenyu Wang, Yufan Zhang, Yuliang Liu 외

The further development of deep neural networks is hampered by the limited GPU memory resource. Therefore, the optimization of GPU memory resources is highly demanded. Swapping and recomputation are commonly applied to m…

GPU

Echo: Compiler-based GPU Memory Footprint Reduction for LSTM RNN Training

2018-05-22 · Bojian Zheng, Abhishek Tiwari, Nandita Vijaykumar, Gennady Pekhimenko

The Long-Short-Term-Memory Recurrent Neural Networks (LSTM RNNs) are a popular class of machine learning models for analyzing sequential data. Their training on modern GPUs, however, is limited by the GPU memory capacity…

GPUMachine TranslationNMT

Reducing Activation Recomputation in Large Transformer Models

2022-05-10 · Vijay Korthikanti, Jared Casper, Sangkug Lym, Lawrence McAfee 외

Training large transformer models is one of the most important computational challenges of modern AI. In this paper, we show how to significantly accelerate training of large transformer models by reducing activation rec…