1-bit Adam: Communication Efficient Large-Scale Training with Adam's Convergence Speed

Abstract

Scalable training of large models (like BERT and GPT-3) requires careful optimization rooted in model design, architecture, and system capabilities. From a system standpoint, communication has become a major bottleneck, especially on commodity systems with standard TCP interconnects that offer limited network bandwidth. Communication compression is an important technique to reduce training time on such systems. One of the most effective methods is error-compensated compression, which offers robust convergence speed even under 1-bit compression. However, state-of-the-art error compensation techniques only work with basic optimizers like SGD and momentum SGD, which are linearly dependent on the gradients. They do not work with non-linear gradient-based optimizers like Adam, which offer state-of-the-art convergence efficiency and accuracy for models like BERT. In this paper, we propose 1-bit Adam that reduces the communication volume by up to $5\times$, offers much better scalability, and provides the same convergence speed as uncompressed Adam. Our key finding is that Adam's variance (non-linear term) becomes stable (after a warmup phase) and can be used as a fixed precondition for the rest of the training (compression phase). Experiments on up to 256 GPUs show that 1-bit Adam enables up to $3.3\times$ higher throughput for BERT-Large pre-training and up to $2.9\times$ higher throughput for SQuAD fine-tuning. In addition, we provide theoretical analysis for our proposed work.

arXiv commentarXiv admin note: text overlap with arXiv:2008.11343

References in corpus (15)

• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• ADADELTA: An Adaptive Learning Rate Method
• TernGrad: Ternary Gradients to Reduce Communication in Distributed Deep Learning
• Sparsified SGD with Memory
• Local SGD Converges Fast and Communicates Little
• Adaptive Gradient Methods with Dynamic Bound of Learning Rate
• PowerSGD: Practical Low-Rank Gradient Compression for Distributed Optimization
• Communication-Computation Efficient Gradient Coding
• DoubleSqueeze: Parallel Stochastic Gradient Descent with Double-Pass Error-Compensated Compression
• Communication-Efficient Distributed Learning via Lazily Aggregated Quantized Gradients
• Linearly Converging Error Compensated SGD
• Double Quantization for Communication-Efficient Distributed Optimization
• PowerGossip: Practical Low-Rank Communication Compression in Decentralized Deep Learning
• Compressing Gradient Optimizers via Count-Sketches
• Adam with Bandit Sampling for Deep Learning

Cited by in corpus (3)

• Large-Scale Deep Learning Optimizations: A Comprehensive Survey
• Layered gradient accumulation and modular pipeline parallelism: fast and efficient training of large language models
• Communication-Compressed Adaptive Gradient Method for Distributed Nonconvex Optimization