Hi,
I looked into the problem and have several solutions -
Problem is that you dont have read-permission for the image.
Solutions are -
I give you the read access. Can give it to you but everytime anyother user uses my script, then will need the permission too. Not a feasible solution.
Export the image to cloud storage and set bucket object permissions to open(Signed URls). But i think it will take longer time(60 GB image) as everytime the system will pull the compressed image and create new instance with it. This might include egress traffic charges also. Again not feasible.
Modify the script to create first instance from marketplace image along with drivers and then make an image from it and from next time onwards use that image. Looks feasible to me.
I will try to modify the script ASAP.
Thanks
@arjundg Thanks for sharing. I am the writer of this azure guide (written as part of previous year course). I have now tried with the steps mentioned above for latest setup. The installation was successful.
Error: module 'fastai.vision.models' has no attribute 'resnet34' Even After I have done conda update -c fastai fastai I get the same error. If I do this models.tvm.resnet34 This solves the problem or resnet but gives me one more error: name 'error_rate' is not defined. What might be the cause of this ? Any solution ?
This worked. After uninstalling anaconda and doing conda update -c fastai fastai was able to install fastai v1.0.11. But why is this the case where we have to uninstall anaconda ?
Was trying notebook outon google collb by following instructions in OP,
got following error.
/usr/local/lib/python3.6/dist-packages/fastai/vision/image.py in open_image(fn)
353 def open_image(fn:PathOrStr)->Image:
354 “Return Image object created from image in file fn.”
–> 355 x = PIL.Image.open(fn).convert(‘RGB’)
356 return Image(pil2tensor(x).float().div_(255))
357
AttributeError: module ‘PIL’ has no attribute ‘Image’
double checked version of fast ai
Requirement already satisfied: fastai in /usr/local/lib/python3.6/dist-packages (1.0.11)
Honestly, it was some over-enthusiastic pinning of versions that we now feel was a mistake. There will be a new release today that reverts this behavior.
Can we have a thread that shows the current version of fastai? It seems as though each nb will require at least version 1.x, it would be nice to know what x is. Also it would be nice if the thread broke it down into conda,pip and git versions. Most of the setup issues seem to have at least something to do with version control.
