Partial Parameter Updates for Efficient Distributed Training
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866912608670449664 |
|---|---|
| author | Filippova, Anastasiia Katharopoulos, Angelos Grangier, David Collobert, Ronan |
| author_facet | Filippova, Anastasiia Katharopoulos, Angelos Grangier, David Collobert, Ronan |
| contents | We introduce a memory- and compute-efficient method for low-communication distributed training. Existing methods reduce communication by performing multiple local updates between infrequent global synchronizations. We demonstrate that their efficiency can be significantly improved by restricting backpropagation: instead of updating all the parameters, each node updates only a fixed subset while keeping the remainder frozen during local steps. This constraint substantially reduces peak memory usage and training FLOPs, while a full forward pass over all parameters eliminates the need for cross-node activation exchange. Experiments on a $1.3$B-parameter language model trained across $32$ nodes show that our method matches the perplexity of prior low-communication approaches under identical token and bandwidth budgets while reducing training FLOPs and peak memory. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_22418 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Partial Parameter Updates for Efficient Distributed Training Filippova, Anastasiia Katharopoulos, Angelos Grangier, David Collobert, Ronan Machine Learning Artificial Intelligence We introduce a memory- and compute-efficient method for low-communication distributed training. Existing methods reduce communication by performing multiple local updates between infrequent global synchronizations. We demonstrate that their efficiency can be significantly improved by restricting backpropagation: instead of updating all the parameters, each node updates only a fixed subset while keeping the remainder frozen during local steps. This constraint substantially reduces peak memory usage and training FLOPs, while a full forward pass over all parameters eliminates the need for cross-node activation exchange. Experiments on a $1.3$B-parameter language model trained across $32$ nodes show that our method matches the perplexity of prior low-communication approaches under identical token and bandwidth budgets while reducing training FLOPs and peak memory. |
| title | Partial Parameter Updates for Efficient Distributed Training |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2509.22418 |