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...
محفوظ في:
| المؤلفون الرئيسيون: | , , |
|---|---|
| التنسيق: | Artigo |
| اللغة: | Inglês |
| منشور في: |
IEEE
2024-01-01
|
| سلاسل: | IEEE Access |
| الموضوعات: | |
| الوصول للمادة أونلاين: | https://ieeexplore.ieee.org/document/10381706/ |
| الوسوم: |
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|