Pipeline Parallelism With Elastic Averaging
To accelerate the training speed of massive DNN models on large-scale datasets, distributed training techniques, including data parallelism and model parallelism, have been extensively studied. In particular, pipeline parallelism, which is derived from model parallelism, has been attracting attentio...
Bewaard in:
| Hoofdauteurs: | , , |
|---|---|
| Formaat: | Artigo |
| Taal: | Inglês |
| Gepubliceerd in: |
IEEE
2024-01-01
|
| Reeks: | IEEE Access |
| Onderwerpen: | |
| Online toegang: | https://ieeexplore.ieee.org/document/10381706/ |
| Tags: |
Geen labels, Wees de eerste die dit record labelt!
|