GPUEater is a cloud computing service focusing on Machine Learning and Deep Learning. Now, AMD Radeon GPUs and NVIDIA Quadro GPUs are available.

This document is intended to describe how to set up this API and how to control your instances through this API.

Before getting started, register your account on GPUEater. https://www.gpueater.com/

• How to setup Radeon GPU Driver (ROCm) on Ubuntu16.04/18.04
• How to setup ROCm-Tensorflow on Ubuntu16.04/18.04
  • ROCm(AMDGPU)-TensorFlow 1.8 Python2.7/Python3.5 + UbuntuOS
  • ROCm(AMDGPU)-TensorFlow 1.1xx Python2.7/Python3.5/Python3.6 + UbuntuOS
  • CPU-TensorFlow 1.10.1 Python3.7 + MacOSX
• Lightweight ROCm-TensorFlow docker
  • ROCm-TensorFlow on GPUEater
  • ROCm-TensorFlow1.11 docker

ROCm information URL : https://github.com/aieater/rocm_tensorflow_info

1. Ruby 2.x is required to run GPUEater API Console.
2. Create a JSON file in accordance with the following instruction.

At first, open your account page(https://www.gpueater.com/console/account) and copy your access_token. The next, create a JSON file on ~/.eater

{
        "gpueater": {
                "access_token":"[YourAccessToken]",
                "secret_token":"[YourSecretToken]"
        }
}

or

{
        "gpueater": {
                "email":"[YourEmail]",
                "password":"[YourPassword]"
        }
}

• At this time, permission control for each token is not available. Still in development.

Add this line to your application's Gemfile:

gem 'gpueater'

And then execute:

$ bundle

Or install it yourself as:

$ gem install gpueater

Before launching an instance, you need to decide product, ssh key, OS image. Get each info with the following APIs.

This API returns current available on-demand products.

require('gpueater')

g = GPUEater.new
puts g.ondemand_list()

This API returns your registered ssh keys.

require('gpueater')

g = GPUEater.new
puts g.ssh_key_list()

This API returns available OS images.

require('gpueater')

g = GPUEater.new
puts g.image_list()

Specify product, OS image, and ssh_key for instance launching.

require('gpueater')

g = GPUEater.new
res = g.ondemand_list()

image = res.find_image('Ubuntu16.04 x64')
ssh_key = res.find_ssh_key('[Your ssh key]')
product = res.find_product('a1.rx580')

param = {
    'product_id' => product['id'],
    'image' => image['alias'],
    'ssh_key_id' => ssh_key['id'],
    'tag' => 'HappyGPUProgramming'
}

res = g.launch_ondemand_instance(param)
puts res

In the event, the request has succeeded, then the API returns the following empty data. {data:null, error:null}

In the event, errors occurred during the instance instantiation process, then the API returns details about the error.

This API returns your launched instance info.

require('gpueater')

g = GPUEater.new
res = g.instance_list()

Before terminating an instance, get instance info through instance list API. Your instance_id and machine_resource_id are needed to terminate.

require('gpueater')

g = GPUEater.new
res = g.instance_list()
res.select{|e| e['tag'] == 'HappyGPUProgramming' }.each{|e|
	puts g.terminate_instance(e)
}

require 'fileutils'
require 'gpueater'

g = GPUEater.new

keyname = 'my_ssh_key2'


# delete old keys.
g.ssh_key_list().select{|e| g.delete_ssh_key(e) if e["name"] == keyname }



# generate key pair and register public key.
key = g.generate_ssh_key
g.register_ssh_key({"name"=>keyname,"public_key"=>key["public_key"]})

# store private key to ~/.ssh/***.pem file.
homedir     = File.expand_path('~')
pem = File.join(homedir,'.ssh',keyname+".pem")
fp = open(pem,"w")
fp.write(key["private_key"])
fp.close

# change mode.
FileUtils.chmod(0600,pem)


puts g.ssh_key_list

The "machine_resource_id" is including an instance object. See the following sample code.

Example:

instance = g.instance_list()[0]
# instance object has instance_id, and machine_resource_id.

g.terminate_instance(instance)

This project is licensed under the MIT License - see the LICENSE.txt file for details