fastai-course-v3/docs/start_vastai.md

title	keywords	sidebar
vast.ai	<nil>	home_sidebar

Welcome to vast.ai beta

https://vast.ai

vasta.ai provides access to GPUs on third party machines. Data may not be private or secure. See discussion here: https://vast.ai/faq/

Pricing

Varies based on supply and demand.

10GB storage is included in the price shown (it's $.002/hr, as you see if bidding for a server)

It costs $.02/GB to upload and download data.

You get $2.00 of credit when you enter your payment info.

Step 1: Navigate your browser to https://vast.ai/console/create/

Step 2: Create an account

Fill out the form and check your email to confirm signup.

Step 3: Choose your image

Click Select Image.Choose pytorch/pytorch. For version tag to use choose the one that starts with nightly_devel_cuda9.2

Step 4: Choose whether to rent or bid.

Accept a price and start by pressing Rent on the machine of your choice. Or make a Make Bid to bid on a machine. If you a make a bid and then someone rents the machine at the full price, you loose out. Renting leads to less frustration than bidding.

If a machine has an AWS, Google Cloud, or Paperspace logo, then clicking Rent will just take you to that provider's website.

Step 5: Enter your payment info.

You will be asked to setup your payment info.

Payment info from vast.ai: Billed every Friday. You won't be charged until your credits run out. We use Stripe , so your card is never sent to our servers.

Step 6: Start instance

It will take a few minutes to load. Click on Instances in the menu on the left. Click on Start on your instance Click on Connect when the button changes to Connect

Step 7: Your JupyterNotebook or JupyterLab should automatically open on your browser at "https://jupyter.vast.ai/..."

Install fastai and you're off!

!pip install fastai

import torch

print(torch.__version__)
print(torch.cuda.is_available())
print(torch.backends.cudnn.enabled)

It should take a minute or two to install fastai. Then you should see output that starts with 1.0 and True True

import fastai
print(fastai.__version__)
from fastai import *
from fastai.vision import *

You should see output that starts with 1.0

Proceed to code.

Don't forget to click stop on your instance at vast.ai -> Instances or you will continue to be billed!