Efficient Stagewise Pretraining via Progressive Subnetworks

• URL: http://arxiv.org/abs/2402.05913v1
• Date: Thu, 8 Feb 2024 18:49:09 GMT
• Title: Efficient Stagewise Pretraining via Progressive Subnetworks
• Authors: Abhishek Panigrahi, Nikunj Saunshi, Kaifeng Lyu, Sobhan Miryoosefi, Sashank Reddi, Satyen Kale, Sanjiv Kumar
• Abstract summary: We propose an alternative framework, progressive subnetwork training, that maintains the full model throughout training, but only trainsworks within the model in each step. RaPTr achieves better pre-training loss for BERT and UL2 language models while requiring 20-33% fewer FLOPs compared to standard training, and is competitive or better than other efficient training methods.
• Score: 55.65819977062729
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recent developments in large language models have sparked interest in efficient pretraining methods. A recent effective paradigm is to perform stage-wise training, where the size of the model is gradually increased over the course of training (e.g. gradual stacking (Reddi et al., 2023)). While the resource and wall-time savings are appealing, it has limitations, particularly the inability to evaluate the full model during earlier stages, and degradation in model quality due to smaller model capacity in the initial stages. In this work, we propose an alternative framework, progressive subnetwork training, that maintains the full model throughout training, but only trains subnetworks within the model in each step. We focus on a simple instantiation of this framework, Random Path Training (RaPTr) that only trains a sub-path of layers in each step, progressively increasing the path lengths in stages. RaPTr achieves better pre-training loss for BERT and UL2 language models while requiring 20-33% fewer FLOPs compared to standard training, and is competitive or better than other efficient training methods. Furthermore, RaPTr shows better downstream performance on UL2, improving QA tasks and SuperGLUE by 1-5% compared to standard training and stacking. Finally, we provide a theoretical basis for RaPTr to justify (a) the increasing complexity of subnetworks in stages, and (b) the stability in loss across stage transitions due to residual connections and layer norm.

Related papers

• Joint or Disjoint: Mixing Training Regimes for Early-Exit Models [3.052154851421859]
  Early exits significantly reduce the amount of computation required in deep neural networks. Most early exit methods employ a training strategy that either simultaneously trains the backbone network and the exit heads or trains the exit heads separately. We propose a training approach where the backbone is initially trained on its own, followed by a phase where both the backbone and the exit heads are trained together.
  arXiv  Detail & Related papers  (2024-07-19T13:56:57Z)
• Boosting Meta-Training with Base Class Information for Few-Shot Learning [35.144099160883606]
  We propose an end-to-end training paradigm consisting of two alternative loops. In the outer loop, we calculate cross entropy loss on the entire training set while updating only the final linear layer. This training paradigm not only converges quickly but also outperforms existing baselines, indicating that information from the overall training set and the meta-learning training paradigm could mutually reinforce one another.
  arXiv  Detail & Related papers  (2024-03-06T05:13:23Z)
• Fast Propagation is Better: Accelerating Single-Step Adversarial Training via Sampling Subnetworks [69.54774045493227]
  A drawback of adversarial training is the computational overhead introduced by the generation of adversarial examples. We propose to exploit the interior building blocks of the model to improve efficiency. Compared with previous methods, our method not only reduces the training cost but also achieves better model robustness.
  arXiv  Detail & Related papers  (2023-10-24T01:36:20Z)
• TWINS: A Fine-Tuning Framework for Improved Transferability of Adversarial Robustness and Generalization [89.54947228958494]
  This paper focuses on the fine-tuning of an adversarially pre-trained model in various classification tasks. We propose a novel statistics-based approach, Two-WIng NormliSation (TWINS) fine-tuning framework. TWINS is shown to be effective on a wide range of image classification datasets in terms of both generalization and robustness.
  arXiv  Detail & Related papers  (2023-03-20T14:12:55Z)
• Learning to Weight Samples for Dynamic Early-exiting Networks [35.03752825893429]
  Early exiting is an effective paradigm for improving the inference efficiency of deep networks. Our work proposes to adopt a weight prediction network to weight the loss of different training samples at each exit. We show that the proposed weighting mechanism consistently improves the trade-off between classification accuracy and inference efficiency.
  arXiv  Detail & Related papers  (2022-09-17T10:46:32Z)
• Layer-Wise Partitioning and Merging for Efficient and Scalable Deep Learning [16.38731019298993]
  We have proposed a novel layer-wise partitioning and merging, forward and backward pass parallel framework to provide better training performance. The experimental evaluation on real use cases shows that the proposed method outperforms the state-of-the-art approaches in terms of training speed.
  arXiv  Detail & Related papers  (2022-07-22T11:47:34Z)
• Simultaneous Training of Partially Masked Neural Networks [67.19481956584465]
  We show that it is possible to train neural networks in such a way that a predefined 'core' subnetwork can be split-off from the trained full network with remarkable good performance. We show that training a Transformer with a low-rank core gives a low-rank model with superior performance than when training the low-rank model alone.
  arXiv  Detail & Related papers  (2021-06-16T15:57:51Z)
• Dynamic Sparse Training for Deep Reinforcement Learning [36.66889208433228]
  We propose for the first time to dynamically train deep reinforcement learning agents with sparse neural networks from scratch. Our approach is easy to be integrated into existing deep reinforcement learning algorithms. We evaluate our approach on OpenAI gym continuous control tasks.
  arXiv  Detail & Related papers  (2021-06-08T09:57:20Z)
• Cream of the Crop: Distilling Prioritized Paths For One-Shot Neural Architecture Search [60.965024145243596]
  One-shot weight sharing methods have recently drawn great attention in neural architecture search due to high efficiency and competitive performance. To alleviate this problem, we present a simple yet effective architecture distillation method. We introduce the concept of prioritized path, which refers to the architecture candidates exhibiting superior performance during training. Since the prioritized paths are changed on the fly depending on their performance and complexity, the final obtained paths are the cream of the crop.
  arXiv  Detail & Related papers  (2020-10-29T17:55:05Z)
• Subset Sampling For Progressive Neural Network Learning [106.12874293597754]
  Progressive Neural Network Learning is a class of algorithms that incrementally construct the network's topology and optimize its parameters based on the training data. We propose to speed up this process by exploiting subsets of training data at each incremental training step. Experimental results in object, scene and face recognition problems demonstrate that the proposed approach speeds up the optimization procedure considerably.
  arXiv  Detail & Related papers  (2020-02-17T18:57:33Z)

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.