Mesa: A Memory-saving Training Framework for Transformers

• URL: http://arxiv.org/abs/2111.11124v1
• Date: Mon, 22 Nov 2021 11:23:01 GMT
• Title: Mesa: A Memory-saving Training Framework for Transformers
• Authors: Zizheng Pan, Peng Chen, Haoyu He, Jing Liu, Jianfei Cai, Bohan Zhuang
• Abstract summary: We present Mesa, a memory-saving training framework for Transformers. Mesa uses exact activations during forward pass while storing a low-precision version of activations to reduce memory consumption during training. Experiments on ImageNet, CIFAR-100 and ADE20K demonstrate that Mesa can reduce half of the memory footprints during training.
• Score: 58.78933015299703
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: There has been an explosion of interest in designing high-performance Transformers. While Transformers have delivered significant performance improvements, training such networks is extremely memory intensive owing to storing all intermediate activations that are needed for gradient computation during backpropagation, especially for long sequences. To this end, we present Mesa, a memory-saving resource-efficient training framework for Transformers. Specifically, Mesa uses exact activations during forward pass while storing a low-precision version of activations to reduce memory consumption during training. The low-precision activations are then dequantized during back-propagation to compute gradients. Besides, to address the heterogeneous activation distributions in the multi-head self-attention layers, we propose a head-wise activation quantization strategy, which quantizes activations based on the statistics of each head to minimize the approximation error. To further boost training efficiency, we learn quantization parameters by running estimates. More importantly, by re-investing the saved memory in employing a larger batch size or scaling up model size, we may further improve the performance under constrained computational resources. Extensive experiments on ImageNet, CIFAR-100 and ADE20K demonstrate that Mesa can reduce half of the memory footprints during training while achieving comparable or even better performance. Code is available at https://github.com/zhuang-group/Mesa

Related papers

• Efficiently Training 7B LLM with 1 Million Sequence Length on 8 GPUs [24.066283519769968]
  Large Language Models (LLMs) have been trained using extended context lengths to foster more creative applications. We propose MEMO, a novel framework for fine-grained activation memory management. We show that MEMO achieves an average of 2.42x and 2.26x MFU compared to Megatron-LM and DeepSpeed.
  arXiv  Detail & Related papers  (2024-07-16T18:59:49Z)
• SHERL: Synthesizing High Accuracy and Efficient Memory for Resource-Limited Transfer Learning [63.93193829913252]
  We propose an innovative METL strategy called SHERL for resource-limited scenarios. In the early route, intermediate outputs are consolidated via an anti-redundancy operation. In the late route, utilizing minimal late pre-trained layers could alleviate the peak demand on memory overhead.
  arXiv  Detail & Related papers  (2024-07-10T10:22:35Z)
• Reducing Fine-Tuning Memory Overhead by Approximate and Memory-Sharing Backpropagation [29.139579820699495]
  This work strives to reduce memory overhead in fine-tuning from perspectives of activation function and layer normalization. We apply our Approx-BP theory to backpropagation training and derive memory-efficient alternatives of GELU and SiLU activation functions. In addition, we introduce a Memory-Sharing Backpropagation strategy, which enables the activation memory to be shared by two adjacent layers.
  arXiv  Detail & Related papers  (2024-06-24T03:09:15Z)
• Block Selective Reprogramming for On-device Training of Vision Transformers [12.118303034660531]
  We present block selective reprogramming (BSR) in which we fine-tune only a fraction of total blocks of a pre-trained model. Compared to the existing alternatives, our approach simultaneously reduces training memory by up to 1.4x and compute cost by up to 2x.
  arXiv  Detail & Related papers  (2024-03-25T08:41:01Z)
• Blockwise Parallel Transformer for Large Context Models [70.97386897478238]
  Blockwise Parallel Transformer (BPT) is a blockwise computation of self-attention and feedforward network fusion to minimize memory costs. By processing longer input sequences while maintaining memory efficiency, BPT enables training sequences 32 times longer than vanilla Transformers and up to 4 times longer than previous memory-efficient methods.
  arXiv  Detail & Related papers  (2023-05-30T19:25:51Z)
• Winner-Take-All Column Row Sampling for Memory Efficient Adaptation of Language Model [89.8764435351222]
  We propose a new family of unbiased estimators called WTA-CRS, for matrix production with reduced variance. Our work provides both theoretical and experimental evidence that, in the context of tuning transformers, our proposed estimators exhibit lower variance compared to existing ones.
  arXiv  Detail & Related papers  (2023-05-24T15:52:08Z)
• Tempo: Accelerating Transformer-Based Model Training through Memory Footprint Reduction [3.5831119917067737]
  We propose Tempo, a new approach to efficiently use accelerator memory resources for training Transformer-based models. Our approach provides drop-in replacements for the GELU, LayerNorm, and Attention layers, reducing the memory usage. We demonstrate that Tempo enables up to 2x higher batch sizes and 16% higher training throughput over the state-of-the-art baseline.
  arXiv  Detail & Related papers  (2022-10-19T01:59:37Z)
• ActNN: Reducing Training Memory Footprint via 2-Bit Activation Compressed Training [68.63354877166756]
  ActNN is a memory-efficient training framework that stores randomly quantized activations for back propagation. ActNN reduces the memory footprint of the activation by 12x, and it enables training with a 6.6x to 14x larger batch size.
  arXiv  Detail & Related papers  (2021-04-29T05:50:54Z)
• Improving Computational Efficiency in Visual Reinforcement Learning via Stored Embeddings [89.63764845984076]
  We present Stored Embeddings for Efficient Reinforcement Learning (SEER) SEER is a simple modification of existing off-policy deep reinforcement learning methods. We show that SEER does not degrade the performance of RLizable agents while significantly saving computation and memory.
  arXiv  Detail & Related papers  (2021-03-04T08:14:10Z)
• Improving compute efficacy frontiers with SliceOut [31.864949424541344]
  We introduce SliceOut -- a dropout-inspired scheme to train deep learning models faster without impacting final test accuracy. At test time, turning off SliceOut performs an implicit ensembling across a linear number of architectures that preserves test accuracy. This leads to faster processing of large computational workloads overall, and significantly reduce the resulting energy consumption and CO2emissions.
  arXiv  Detail & Related papers  (2020-07-21T15:59:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.