How to run experiments on multiple GPUs environment using ibmfl package?

IBM / federated-learning-lib

A library for federated learning (a distributed machine learning process) in an enterprise environment.

Other

500 stars 137 forks source link

How to run experiments on multiple GPUs environment using ibmfl package? #34

Closed moFang222 closed 3 years ago

moFang222 commented 4 years ago

Hi. Follow the tutorial, I can run the program under the cpu environment. But how can I run the program under the gpu environment? Is it simply done by installing the tensorflow-gpu or does it need other setup? Many thanks!

Yi-Zoey commented 4 years ago

Hi, thanks for reaching out! If you want to train a Keras model with only 1 GPU, you can just install the corresponding tensorflow-gpu package. We will be supporting multiple GPUs in our coming release. Stay tuned!

Yi-Zoey commented 4 years ago

Hi, please check out our new release with multiple GPU support and the corresponding tutorial for configure gpu training.

saigontrade88 commented 2 years ago

Hi, for pytorch models, using single-process DataParallel with multiple GPUs is slow. Pytorch does suggest using DistributedDataParallel for multiple GPU training in this link. Can you suggest how to implement multiple GPU training? Many thanks.