# LAYUP: Asynchronous decentralized gradient descent with LAYer-wise UPdates

arXiv: https://arxiv.org/abs/2410.05985 · HTML: https://mlsanity.com/paper/2410.05985/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 23 of its 45 references that are in the map.
- First posted on arXiv 8 Oct 2024.

## Builds on (most cited first, 23 of 23)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [PaLM: Scaling Language Modeling with Pathways](https://mlsanity.com/paper/2204.02311/) (2022)
- [Pointer Sentinel Mixture Models](https://mlsanity.com/paper/1609.07843/) (2016)
- [ZeRO: Memory Optimizations Toward Training Trillion Parameter Models](https://mlsanity.com/paper/1910.02054/) (2019)
- [PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel](https://mlsanity.com/paper/2304.11277/) (2023)
- [Local SGD Converges Fast and Communicates Little](https://mlsanity.com/paper/1805.09767/) (2018)
- [Don't Use Large Mini-Batches, Use Local SGD](https://mlsanity.com/paper/1808.07217/) (2018)
- [Asynchronous Decentralized Parallel Stochastic Gradient Descent](https://mlsanity.com/paper/1710.06952/) (2017)
- [Stochastic Gradient Push for Distributed Deep Learning](https://mlsanity.com/paper/1811.10792/) (2018)
- [Is Local SGD Better than Minibatch SGD?](https://mlsanity.com/paper/2002.07839/) (2020)
- [SlowMo: Improving Communication-Efficient Distributed SGD with Slow Momentum](https://mlsanity.com/paper/1910.00643/) (2019)
- [Asynchronous Stochastic Gradient Descent with Delay Compensation](https://mlsanity.com/paper/1609.08326/) (2016)
- [The MiniPile Challenge for Data-Efficient Language Models](https://mlsanity.com/paper/2304.08442/) (2023)
- [Asynchronous SGD Beats Minibatch SGD Under Arbitrary Delays](https://mlsanity.com/paper/2206.07638/) (2022)
- [Accelerating Gossip SGD with Periodic Global Averaging](https://mlsanity.com/paper/2105.09080/) (2021)
- [Asynchronous Gradient-Push](https://mlsanity.com/paper/1803.08950/) (2018)
- [CO2: Efficient Distributed Training with Full Communication-Computation Overlap](https://mlsanity.com/paper/2401.16265/) (2024)
- [Asynchronous SGD on Graphs: a Unified Framework for Asynchronous Decentralized and Federated Optimization](https://mlsanity.com/paper/2311.00465/) (2023)
- [Distributed Training of Deep Neural Networks: Theoretical and Practical Limits of Parallel Scalability](https://mlsanity.com/paper/1609.06870/) (2016)
- [Doing More by Doing Less: How Structured Partial Backpropagation Improves Deep Learning Clusters](https://mlsanity.com/paper/2111.10672/) (2021)
- [Scaling the Wild: Decentralizing Hogwild!-style Shared-memory SGD](https://mlsanity.com/paper/2203.06638/) (2022)

## Abstract (arXiv)

The increasing size of deep learning models has made distributed training across multiple devices essential. Synchronous, centralized methods incur large communication and synchronization overheads. Communication efficient algorithms can reduce these overheads, but often require extra buffers, remain sensitive to stragglers or parameter drift. We present LayUp, an asynchronous decentralized SGD method with layer-wise updates. LayUp asynchronously exchanges incremental layer-wise updates during backpropagation. It uses randomized gossip communication, enabling updates to be applied as soon as they are available without buffering. These design choices reduce parameter drift and improve robustness to stragglers. We establish a theoretical upper bound for the gradient bias introduced by layer-wise updates and prove convergence of LayUp. We empirically validate LayUp on vision and language modeling tasks, showing convergence up to ~32% faster in terms of wall-clock time compared to synchronous data parallel training and up to ~27% faster than comparable communication efficient algorithms while maintaining better task performance. This speed-up is partly due to higher model FLOPs utilization, as we demonstrate. By injecting delays into the communication between workers, we show that LayUp remains robust to stragglers while DDP and other methods degrade in performance. Overall, LayUp provides a novel practical, straggler-robust alternative for distributed training without sacrificing accuracy.