Amazing! Yes I have an uppercase in my username. Added a new user and I’m in.
Thanks for your help.
I’ve managed to setup a regular .90 per hour VM, and then realized it was a different template link for the .18
The limit is 10 cores with basic pay-as-you-go, so instead of deleting my existing VM, I submitted a support request to bump it from 10 to 12, so I could try the .18 and keep my existing. Not sure I did that right. It was approved, but still get the error about only having 10 cores. 
Question for you: Is the .18 working for you? Has it been removed yet? Wondering if it’s worth the hassle. Not sure I really understand how challenging this statement is: “Your work from the deallocated VM is not saved by default.”
Thanks
I have been using the preempt-able version for weeks. Never lost anything. I do download my files at the end of the day to be safe, but I would do this regardless which VM I chose.
One piece of advice, choose East US as the resource group region for the machine. This machine is only offered in a few select regions. The errors are hard to understand if you select the wrong region.
1 Like
Thanks for the tip and replying, I’ve been having issues with what location offered what service.
I think this weekend, I’ll clean up my mess and go with .18 on East US.
I get an error microsoft-dsvm:linux-data-science-vm-ubuntu:linuxdsvmubuntu:18.12.01 is not available when I try to follow the tutorial using the template https://course.fast.ai/start_azure.html
Does the template need to be changed?
Many thanks,
Chintan
The templates in the tutorial do not work for me. Both of them lead to the error ‘resource not available…’ after clicking ‘purchase’. I tried out different regions, and have sufficient credits available.
Error message: ‘InvalidTemplateDeployment’.
I now created a custom NC6 instance instead (which is more costly), but of course I don’t have fastai in jupyter now. Could someone please fix the templates.
In the regular (non preemptable) Data Science VM (DSVM) instance template, we previously had version pinned to “18.12.01”. I just fixed it to pickup the latest verion of the DSVM image. The preemptable instance was already using the latest available version of the DSVM image.
@BelAir - On the portal deployment there will be more detailed error message for the InvalidTemplateDeployment. Please share the same. Were you using the regular instance or preemptable instances? You can also retry if you were using a regular Data Science VM instance.
1 Like
The template was changed to pick the latest available version of Data Science VM image instead of hardcoded version (18.12.01) earlier. If you do a new deployment it will automatically pick the latest version.
1 Like
thanks @zenlytix I can confirm that the deployment is working now.
1 Like
The error message is not helpful:
The template deployment failed with error: ‘The resource with id: … failed validation with message: ‘The requested size for resource … is currently not available in location ‘westus2’ zones ‘’ for subscription … Please try another size or deploy to a different location or zones. See https://aka.ms/azureskunotavailable for details.’.’. (Code: InvalidTemplateDeployment).
I am using an Azure account with a student subscription. Maybe that is the issue.
Usually that means the requested VM size is not available in the region at that time when you requested it. What kind of VM size are you trying to create and which region? Also is this a low priority (preempatable) or regular instance?
I had an incredible amount of trouble with this myself. The issue in the end was that I was using an “NC6 Promo” processor.
I resolved by completing the template fields (as per the fast.ai documentation link) and then editing the template to manually change the value to NC6 Promo rather than just NC6. Once the template is edited, I deployed and it worked as expected.
I am guessing the standard template was throwing the error because there was a promo on the processor and the standard NC6 not available (ie. whilst the promo was on).
2 Likes
The NC6 Promo is special VM SKU available for a limited time. The template used the regular NC6 (either dedicated or low priority instances). NC6 promo costs about $0.40. Low priority is still cheaper but of course there is a chance it will get preempted. I will update the script to use NCx Promo for the dedicated instances and leave it in as long it is available. In the meantime @scotth procedure above is a reasonable workaround (Thanks for trying this and sorry about the trouble. If you can writeup the specific issues you faced, I will be sure to pass it along to folks in Azure).
Yes I am so sorry, I should have been clearer. I was getting this type of error but in aueast
I tried to start a preemptible VM with V100 using the link in the tutorial, but the created instance does not seem to have a public IP (and hence no Connect button). Is there anything I am missing?
Preemptible instances have a load balancer and VM scale set. You have to use the IP for the VM scale set. Here is note from Azure docs page for fast.ai for IP information:
"If you created a pre-emptable instance, it will appear on Azure portal as a “Virtual machine scale set” instead of a “Virtual Machine”. ".
Sorry for the trouble. Ideally this should have been more easily visible.
1 Like
Thank you - it all works now.
@zenlytix: Tried creating linux DSVM from the fast.ai template, still the same error( instance not available). As Scott mentioned edited to NC6_Promo and it worked.
1 Like