AWS AMI available for testing


(Jeremy Howard) #1

For folks who are confident in getting around AWS and Linux, please test out this AMI I just created - name is fastai-part1v2-p2, id is ami-8c4288f4, region is Oregon. NB: this only works with P2 instances.

The repo is in ~/fastai, and all libs and the dogscats data are installed and ready to go. So you should be able to run lesson one without changes - although I’d suggest doing a git pull first.

Any feedback or comments welcomed!


How to set up AWS for Part 1 v2
AWS GPU install script and public AMI
(Dominique MIAS) #2

@jeremy Thank you very much ! Is there any way to test it in another region ?


(Ankit Goila) #3

@jeremy I was able to create a p2 instance and ssh into it using this new AMI.
I’m not able to run a jupyter notebook however. Don’t know what’s going wrong.

For the instance, I just typed: jupyter notebook

In my browser, when I do: <my-elastic-ip>:8888/?token=.... , it just displays a message Refused to connect.


(yinterian) #4

Did you try tunneling?

Also you may need to configure jupyter notebook.

Configuring jupyter notebook

jupyter notebook --generate-config

To create a password run

#jupyter notebook password
echo “c.NotebookApp.ip = '*'
c.NotebookApp.open_browser = False” >> $HOME/.jupyter/jupyter_notebook_config.py


#5

@jeremy Is it possible to put this new AMI to the region of Ohio as well? Not able to try due to gpu limits in Oregon. Thanks


#6

Jupyter notebook works fine with this new AMI, ami-5f2fe427. Did you do tunneling when you ssh to the instance? If not, you need to. The command:

$ ssh -i <path_to_your_.ssh>/<your_key_file.pem> ubuntu@<your_instance_ip> -L 8888:127.0.0.1:8888


(Ankit Goila) #7

Interesting. I did do the ssh tunneling. I’ll try that setup again. Thanks!


#8

@jeremy
For AWS ami-5f2fe427, lesson1.ipynb, the following needs update. The data is there, just needs to point the path to it, do:

PATH = “/home/ubuntu/data/dogscats/”

no need for:

Extra steps if NOT using Crestle (e.g. if you’re using AWS …)
The dataset is available at http://files.fast.ai/data/dogscats.zip. You can download it directly on your server by running the following line in your terminal. wget http://files.fast.ai/data/dogscats.zip. You should put the data in a subdirectory of this notebook’s directory, called data/


(Jeremy Howard) #9

It’s already pointing at it - there’s a data symlink in the spot PATH points to, I believe.


(Benjamin DeKoven) #10

I am using PuTTY. Launching Jupyter Notebook also worked fine. The path to the data needed no modification.

Here are some results from testing the new AMI.




(Jeremy Howard) #11

Try rebooting your instance.


(Benjamin DeKoven) #12

I am still having the same issue after a reboot. Then I stopped / restarted and still having the issue.


(Jeremy Howard) #13

Try git pull?


(Benjamin DeKoven) #14

I tried and there are some unmerged files. I will work on this and get back to you sometime on Friday (US date).


(Clayton Yochum) #15

Worked great for me!

Mild hiccup: environment.yml in the repo was modified, so I had to

cd fastai
git checkout -- environment.yml
git pull

to update the repo.

Maybe this is affecting @bdekoven?


(Vikrant Behal) #16

Can we follow instructions from V1 to try it?


(Arjun Rajkumar) #17

Works well. Didn’t have to git pull. Thanks!


(Benjamin DeKoven) #18

yes @claytonjy , I had this issue with environment.yml

After fixing “git pull” indicated already up to date. But still have runtime error


(Clayton Yochum) #19

What kind of AWS instance did you choose? I chose p2.xlarge; I think any p2.* should work. Did you choose a non-GPU instance? Could explain the error coming from CUDA.


AWS GPU install script and public AMI
(Jeremy Howard) #20

Yeah good point - anything other than p2 won’t work.