Platform: Google Cloud Platform (GCP)

ilovescience · August 27, 2020, 5:15am

GCP is being more strict about this kind of stuff. Try to have fairly conservative requests and also maybe comment that you are using for fastai course. If they deny it, try to follow up with them as well.

deep-learner · August 27, 2020, 6:36pm

Update: I followed up with them and told them that it’s not possible to get a quota increase for an individual. They said it has to be for a company account (not a gmail account) with a 12 month minimum project. Can anybody else try this out and share their experience please? This seems very weird to me.

micstan · August 27, 2020, 7:21pm

@deep-learner what is your quota request?

ilovescience · August 27, 2020, 8:08pm

So are you unable to get any access to GPUs now?

deep-learner · August 27, 2020, 8:26pm

I followed these instructions and requested to increase “GPUs (all regions)” from 0 to 1. Request got denied, then I was told that ALL quota requests coming from gmail accounts will get denied (sales rep may have been wrong, but that’s what I was told).

deep-learner · August 27, 2020, 8:27pm

I am curious to see if any other new users are able to get a GPU. Mine is just one incident. We should wait to see some more reports. If this trend continues, maybe it’s time to stop recommending GCP for the fastai course.

micstan · August 27, 2020, 8:36pm

For 0->1 request there is definitely something wrong, there are plenty of people using it on private accounts, i haven’t requested recently but it is hard to believe that “ALL quota requests coming from gmail accounts will get denied”.

deep-learner · August 28, 2020, 1:18am

I switched over to Azure for now, because they allow 1 GPU without any quota increase.

jbkjr · August 28, 2020, 8:59pm

I’ve gotten set up with the new course and library on GCP’s AI notebook platform using the instructions on the course homepage provided by @micstan, but since I’ve installed the library with anaconda according to the instructions, I cannot see the lesson notebooks or library files anywhere on my VM (just fastbook and tutorials folders, which contains the old course files).

So, I tried cloning the new repository and renaming the old repository fastai.old as @markphillips did above, but when I go to run the line:
pip install -e “fastai[dev]”

I get the error:

ERROR: “fastai[dev]” is not a valid editable requirement. It should either be a path to a local project or a VCS URL (beginning with svn+, git+, hg+, or bzr+).

Any help on this matter would be greatly appreciated.

micstan · August 29, 2020, 1:43pm

@jbkjr for the lessons notebooks - maybe i miss something but i believe fastbook = lesson notebooks. Since you mentioned you have the directory i believe you successfully downloaded the repo. For the library can you give:

import fastai
print(fastai.__version__)
print(fastai.__file__)

jbkjr · August 30, 2020, 10:05pm

Here’s what I get:
2.0.0
/opt/conda/lib/python3.7/site-packages/fastai/init.py

So, it seems fine? Still, if I wanted to do an editable install of the library in the GCP notebook environment to be able to contribute to its development, how could I do so?

aditya.swami · August 31, 2020, 5:55am

Hi @mcistan,

I tried increasing the quota on gcp to 1 gpu for all regions. The UI has changed a bit so I needed to increase the GPU quota for each region individually.

I then tried to create a notebook in AI Platform but got this error.

Any help will be much appreciated!

Thanks,
Adi

micstan · August 31, 2020, 6:23am

@aditya.swami believe we miss quota request part in the docs. I don’t think you have to increase for each region individually, just all regions, can you follow this:

micstan · August 31, 2020, 6:29am

@jbkjr yes, it looks fine. If you want to contribute check out the docs: https://docs.fast.ai/dev-setup. If you just want to look at the code and is more comfortable in a different location you can just git clone the repo.

aditya.swami · August 31, 2020, 6:52am

Hey @micstan!

Thanks for this. I think the UI has changed a bit since that post.

Or am I missing something?

Thanks!
Adi

micstan · August 31, 2020, 9:05am

Can you find “GPUs (all regions)”? . I will add it later to the docs.

aditya.swami · August 31, 2020, 9:34am

Ah brilliant! That worked! Thanks again for your help @micstan!

mrgold · September 18, 2020, 5:59am

Hi,
I’m in Step 4: Install libraries CUDA update
I’ve downloaded successfully the file and when trying to run this command

sudo sh cuda_10.2.89_440.33.01_linux.run

Prompt is asking for password which I can’t figure what it is and from where I should get it
any ideas?

methodmatters · September 18, 2020, 11:30am

Just following these instructions, and noticed that something has changed in Step 4.

It looks like the "requirements.txt" file is no longer in the course-v4 folder (the cloned directories have been modified on Github recently). For more information, see this note on Github.

The requirements file is now contained in the fastbook directory. Navigate to that folder and then use this code to install the dependencies: pip install -r requirements.txt

micstan · September 19, 2020, 8:38pm

@mrgold are you the owner of the project in GC or maybe using it on your company profile? Unfortunately driver installation has to be done as root. You can try to check with sudo -v and potentially try to change the password with sudo passwd.