No, I get this as error after it says connection is close. Hence, the doubt, why does it consider that as an error?
Thanks a lot for doing this. I received my credits today.
Can we use Git bash in Windows instead of Ubuntu WSL ? I have both but just wondering if Git bash can be used.
I am trying to run keras and tensorflow on GCP to test the speed increase compared to CPU but i have errors installing tensorflow-gpu following https://www.tensorflow.org/install/pip.
In my-fastai-instance, i created and activated virtual env with
virtualenv --system-site-packages -p python3 ./venv
source ./venv/bin/activate
Then pip install --upgrade tensorflow-gpu
i get Failed to load the native TensorFlow runtime.
when verifying installation with
python -c "import tensorflow as tf; print(tf.__version__)"
It cannot import tensorflow.
This does not happen when i pip install tensorflow without GPU support. (it prints versoin 1.11 properly)
During tensorflow-gpu install i see red words fastai 1.0.11 requires torchvision-nightly, which is not installed. Is this an issue?
On another note, why does conda install tensorflow (without GPU support) fail too with PermissionError: [Errno 13] Permission denied: '/opt/anaconda3/.condatmp’
Could someone advice how i can use GPU enabled tensorflow in my-fastai-instance (default one following lesson instructions)
Hey, this is a permission issue. You need to provide write access to /opt directory to the current user.
sudo chmod u+w -R /opt
should resolve this.
No idea about this error , but why dont you create a new conda environment and install tensorflow there using conda. This would be handy like you just need to click on change kernel.
Thanks a lot Petr. I got the extra credits today. Really appreciate your gesture. Thanks to GCP too
Thank you, sir. We all appreciate it.
Idk why, I needed to add another step to those outlined in your blog post. I completed step 3 by pasting the provided code and then typed in http://external-ip:8888
, where external-ip
was the one listed in the terminal, but it wouldn’t connect.
Eventually, I realized it was as if the Jupyter NB server wasn’t running. So, I SSH’d into the new instance the same way you specified for the setup VM and typed jupyter notebook
. It told me it couldn’t find a browser, which is fine b/c when I re-loaded the address with the external ip, I was greeted with the password prompt for the Jupyter notebook.
Between your guide and another $500 worth of free credits, I’m set for a while for virtual machines.
Thanks for your contribution.
Yes it’ll take a time to start jupyter. I’ll note that.
Thanks for your reply. Can you please explain how to do that in google console, some command or screenshot?
@czechcheck Thank you so much for your generosity and your time, you rock! Just got my credit today, I really appreciate it.
Děkuji
@czechcheck Just got the GCP credits, thank you very much and to google for allowing employees to give out perks like this.
I am getting the same error on win 7
Hi,
I am getting the below error when I try to set up boot disk-
ERROR: (gcloud.beta.compute.instances.create) Could not fetch resource:
- Quota ‘GPUS_ALL_REGIONS’ exceeded. Limit: 0.0 globally.
Any solution for this error ?
I am on win 7…have downloaded gc sdk
Not sure about this. May be something related to some billing issue.
Hope someone else could help on this.
From a simple google search it seems like some quota issue.
Check this: https://cloud.google.com/compute/quotas#cpus
I guess no GPU instances are available globally for GCP. I tried to create an instance via console and it still gives me the same error.
Hmm. That’s weird.
Is there a way to install collapsable heading extensions in GCP ?? I tried the usual way of installing through conda forge, I cant seem to get the extra tab that i usually get.
Ik could not get rid of the error so I did my setup using an Ubuntu machine, which ran the step 1 - 4 commands smoothly, with no error.
I now can use my Google Cloud SDK shell in W7 to connect to the instance. No errors since!