FloydHub

FloydHub 是一个深度学习云平台,基于容器技术提供了一个简单易用的深度学习云环境。注册用户免费赠送 2 小时 GPU 和每月 20 小时的 CPU 计算资源以及 10 GB 的存储空间。

安装

$ pip install -U floyd-cli
$ floyd login
Authentication token page will now open in your browser. Continue? [Y/n]: y
Please copy and paste the authentication token.
This is an invisible field. Paste token and press ENTER:

入门示例

首先到 https://www.floydhub.com/projects/create 创建一个项目 mnist-cnn,然后执行

$ git clone https://github.com/floydhub/quick-start.git
Cloning into 'quick-start'...
...
$ cd quick-start

$ floyd init mnist-cnn
Project "mnist-cnn" initialized in current directory

$ floyd run --gpu --env tensorflow-1.3 "python train_and_eval.py"

Creating project run. Total upload size: 25.4KiB
Syncing code ...
[================================] 27316/27316 - 00:00:00

JOB NAME
----------------------
mckay/projects/mnist-cnn/1

To view logs enter:
   floyd logs mckay/projects/mnist-cnn/1

这会

如果使用了 Jupyter Notebook,则可以

$ floyd run --mode jupyter

Creating project run. Total upload size: 198.0B
Syncing code ...
[================================] 946/946 - 00:00:00

JOB NAME
-----------------------------------
mckay/projects/my_jupyter_project/1

Setting up your instance and waiting for Jupyter notebook to become available .............

Path to jupyter notebook: https://floydlabs.com/notebooks/gaftzXTdaPtQtQ9NvEieNg

这会自动在浏览器中打开云端开启的 Jupyter 服务器。

功能特性

  $ floyd run \
  --gpu \
  --env tensorflow-1.3 \
  --data mckay/datasets/mnist/1:/mnist \
  --data mckay/projects/quick-start/1/output:/model \
  'python eval.py'

附注