Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2404.00438 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916186389741568 |
|---|---|
| author | Liu, Bo Wu, Lemeng Chen, Lizhang Liang, Kaizhao Zhu, Jiaxu Liang, Chen Krishnamoorthi, Raghuraman Liu, Qiang |
| author_facet | Liu, Bo Wu, Lemeng Chen, Lizhang Liang, Kaizhao Zhu, Jiaxu Liang, Chen Krishnamoorthi, Raghuraman Liu, Qiang |
| contents | The Lion optimizer has been a promising competitor with the AdamW for training large AI models, with advantages on memory, computation, and sample efficiency. In this paper, we introduce Distributed Lion, an innovative adaptation of Lion for distributed training environments. Leveraging the sign operator in Lion, our Distributed Lion only requires communicating binary or lower-precision vectors between workers to the center server, significantly reducing the communication cost. Our theoretical analysis confirms Distributed Lion's convergence properties. Empirical results demonstrate its robustness across a range of tasks, worker counts, and batch sizes, on both vision and language problems. Notably, Distributed Lion attains comparable performance to standard Lion or AdamW optimizers applied on aggregated gradients, but with significantly reduced communication bandwidth. This feature is particularly advantageous for training large models. In addition, we also demonstrate that Distributed Lion presents a more favorable performance-bandwidth balance compared to existing efficient distributed methods such as deep gradient compression and ternary gradients. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2404_00438 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Communication Efficient Distributed Training with Distributed Lion Liu, Bo Wu, Lemeng Chen, Lizhang Liang, Kaizhao Zhu, Jiaxu Liang, Chen Krishnamoorthi, Raghuraman Liu, Qiang Distributed, Parallel, and Cluster Computing Artificial Intelligence Machine Learning Optimization and Control The Lion optimizer has been a promising competitor with the AdamW for training large AI models, with advantages on memory, computation, and sample efficiency. In this paper, we introduce Distributed Lion, an innovative adaptation of Lion for distributed training environments. Leveraging the sign operator in Lion, our Distributed Lion only requires communicating binary or lower-precision vectors between workers to the center server, significantly reducing the communication cost. Our theoretical analysis confirms Distributed Lion's convergence properties. Empirical results demonstrate its robustness across a range of tasks, worker counts, and batch sizes, on both vision and language problems. Notably, Distributed Lion attains comparable performance to standard Lion or AdamW optimizers applied on aggregated gradients, but with significantly reduced communication bandwidth. This feature is particularly advantageous for training large models. In addition, we also demonstrate that Distributed Lion presents a more favorable performance-bandwidth balance compared to existing efficient distributed methods such as deep gradient compression and ternary gradients. |
| title | Communication Efficient Distributed Training with Distributed Lion |
| topic | Distributed, Parallel, and Cluster Computing Artificial Intelligence Machine Learning Optimization and Control |
| url | https://arxiv.org/abs/2404.00438 |