https://github.com/kentaroy47/pytorch-mgpu-cifar10

testing multi gpu for pytorch
https://github.com/kentaroy47/pytorch-mgpu-cifar10

cifar10 dnn gpu multi-gpu pytorch

Last synced: about 14 hours ago
JSON representation

testing multi gpu for pytorch

• Host: GitHub
• URL: https://github.com/kentaroy47/pytorch-mgpu-cifar10
• Owner: kentaroy47
• Created: 2018-11-28T01:56:42.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2019-05-07T08:58:29.000Z (over 5 years ago)
• Last Synced: 2023-03-04T01:41:50.557Z (almost 2 years ago)
• Topics: cifar10, dnn, gpu, multi-gpu, pytorch
• Language: Python
• Homepage: https://qiita.com/arutema47/items/2b92f94c734b0a11609d
• Size: 14.6 KB
• Stars: 23
• Watchers: 2
• Forks: 8
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Multi-GPU training with Pytorch.

Let's try easy multi-GPU training with pytorch.

Just adding:

```python

if device == 'cuda':

    net = torch.nn.DataParallel(net) # make parallel

    cudnn.benchmark = True

```

will enable parallel GPU usage in Pytorch! :)

Training dataset: CIFAR10

DeepLearningFramework: Pytorch

Should have more than 2GPUs to test it out.

Should scale more than 2 GPUs!

## 詳しい解説と設定など

https://qiita.com/arutema47/items/2b92f94c734b0a11609d

## Usage

```shell

git clone https://github.com/kentaroy47/pytorch-mgpu-cifar10.git

cd pytorch-mgpu-cifar10

export CUDA_VISIBLE_DEVICES=0,1 # parallel training with GPUs 0 and 1.

python train_cifar10.py

# parallel training with GPUs 0-3.

export CUDA_VISIBLE_DEVICES=0,1,2,3

python train_cifar10.py

# デフォルトの設定はResnet18となっています。

# Res101で学習する場合

python train_cifar10.py --net res101

# Res50で学習する場合

python train_cifar10.py --net res50

```