Deconstructing What Makes a Good Optimizer for Language Models

• URL: http://arxiv.org/abs/2407.07972v1
• Date: Wed, 10 Jul 2024 18:11:40 GMT
• Title: Deconstructing What Makes a Good Optimizer for Language Models
• Authors: Rosie Zhao, Depen Morwani, David Brandfonbrener, Nikhil Vyas, Sham Kakade,
• Abstract summary: We compare several optimization algorithms, including SGD, Adafactor, Adam, and Lion, in the context of autoregressive language modeling. Our findings indicate that, except for SGD, these algorithms all perform comparably both in their optimal performance.
• Score: 7.9224468703944115
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Training language models becomes increasingly expensive with scale, prompting numerous attempts to improve optimization efficiency. Despite these efforts, the Adam optimizer remains the most widely used, due to a prevailing view that it is the most effective approach. We aim to compare several optimization algorithms, including SGD, Adafactor, Adam, and Lion, in the context of autoregressive language modeling across a range of model sizes, hyperparameters, and architecture variants. Our findings indicate that, except for SGD, these algorithms all perform comparably both in their optimal performance and also in terms of how they fare across a wide range of hyperparameter choices. Our results suggest to practitioners that the choice of optimizer can be guided by practical considerations like memory constraints and ease of implementation, as no single algorithm emerged as a clear winner in terms of performance or stability to hyperparameter misspecification. Given our findings, we further dissect these approaches, examining two simplified versions of Adam: a) signed momentum (Signum) which we see recovers both the performance and hyperparameter stability of Adam and b) Adalayer, a layerwise variant of Adam which we introduce to study Adam's preconditioning. Examining Adalayer leads us to the conclusion that the largest impact of Adam's preconditioning is restricted to the last layer and LayerNorm parameters, and, perhaps surprisingly, the remaining layers can be trained with SGD.

Related papers

• MARS: Unleashing the Power of Variance Reduction for Training Large Models [56.47014540413659]
  Large gradient algorithms like Adam, Adam, and their variants have been central to the development of this type of training. We propose a framework that reconciles preconditioned gradient optimization methods with variance reduction via a scaled momentum technique.
  arXiv  Detail & Related papers  (2024-11-15T18:57:39Z)
• Adaptive Preference Scaling for Reinforcement Learning with Human Feedback [103.36048042664768]
  Reinforcement learning from human feedback (RLHF) is a prevalent approach to align AI systems with human values. We propose a novel adaptive preference loss, underpinned by distributionally robust optimization (DRO) Our method is versatile and can be readily adapted to various preference optimization frameworks.
  arXiv  Detail & Related papers  (2024-06-04T20:33:22Z)
• Adam with model exponential moving average is effective for nonconvex optimization [45.242009309234305]
  We offer a theoretical analysis of two modern optimization techniques for training large and complex models: (i) adaptive optimization algorithms as Adam, and (ii) the exponential moving average (EMA) model.
  arXiv  Detail & Related papers  (2024-05-28T14:08:04Z)
• Surge Phenomenon in Optimal Learning Rate and Batch Size Scaling [27.058009599819012]
  We study the connection between optimal learning rates and batch sizes for Adam styles. We prove that the optimal learning rate first rises and then falls as the batch size increases.
  arXiv  Detail & Related papers  (2024-05-23T13:52:36Z)
• Variational Stochastic Gradient Descent for Deep Neural Networks [16.96187187108041]
  Current state-of-the-arts are adaptive gradient-based optimization methods such as Adam. Here, we propose to combine both approaches, resulting in the Variational Gradient Descent (VSGD) We show how our VSGD method relates to other adaptive gradient-baseds like Adam.
  arXiv  Detail & Related papers  (2024-04-09T18:02:01Z)
• MADA: Meta-Adaptive Optimizers through hyper-gradient Descent [73.1383658672682]
  We introduce Meta-Adaptives (MADA), a unified framework that can generalize several known convergences and dynamically learn the most suitable one during training. We empirically compare MADA to other populars on vision and language tasks, and find that MADA consistently outperforms Adam and other populars. We also propose AVGrad, a modification of AMS that replaces the maximum operator with averaging, which is more suitable for hyper-gradient optimization.
  arXiv  Detail & Related papers  (2024-01-17T00:16:46Z)
• Studying K-FAC Heuristics by Viewing Adam through a Second-Order Lens [34.72514951778262]
  We study AdamQLR: an optimiser combining damping and learning rate selection techniques from K-FAC. We evaluate AdamQLR on a range of regression and classification tasks at various scales. Finding an untuned AdamQLR setting can achieve comparable performance vs runtime to tuned benchmarks.
  arXiv  Detail & Related papers  (2023-10-23T14:06:46Z)
• Bidirectional Looking with A Novel Double Exponential Moving Average to Adaptive and Non-adaptive Momentum Optimizers [109.52244418498974]
  We propose a novel textscAdmeta (textbfADouble exponential textbfMov averagtextbfE textbfAdaptive and non-adaptive momentum) framework. We provide two implementations, textscAdmetaR and textscAdmetaS, the former based on RAdam and the latter based on SGDM.
  arXiv  Detail & Related papers  (2023-07-02T18:16:06Z)
• MaxVA: Fast Adaptation of Step Sizes by Maximizing Observed Variance of Gradients [112.00379151834242]
  We propose adaptive learning rate principle, in which the running mean of squared gradient in Adam is replaced by a weighted mean, with weights chosen to maximize the estimated variance each coordinate. This results in faster adaptation, which leads more desirable empirical convergence behaviors.
  arXiv  Detail & Related papers  (2020-06-21T21:47:43Z)
• ADAHESSIAN: An Adaptive Second Order Optimizer for Machine Learning [91.13797346047984]
  We introduce ADAHESSIAN, a second order optimization algorithm which dynamically incorporates the curvature of the loss function via ADAptive estimates. We show that ADAHESSIAN achieves new state-of-the-art results by a large margin as compared to other adaptive optimization methods.
  arXiv  Detail & Related papers  (2020-06-01T05:00:51Z)

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.