Uploadinig directly would be a huge help. I can’t make google drive not display the ‘no preview’ and am pretty sure that is what is blocking the download and thus pickle fails b/c it’s loading html and not a file.
I can download/view the pkl file fine on my hard drive so no issue there.
When I go into the render interface I don’t see any option to upload things? Do I have to put it in my github and then it pulls or how to do this? That should be the fix to get this running.
Most likely not. It’s no preview due to it just doesn’t know what to do with that type of file extension. What are the sharing settings on the file itself?
re: sharing - I have it set as “anyone with the link can access, no sign in required”.
Hi LessW2020
Here is an old classifier that classifies English pound notes £5, £10, £20
https://drive.google.com/uc?export=download&id=12ak8K6dFwVUgKmDLRUPgwt6vB80kr2Cy
Does it open correctly?
cheers mrfabulous1 
HIi @mrfabulous1 - yes that opens as expected (direct to drive).
However, even when I changed my google drive to a ‘download link’ via rawdownload.now.sh, that then generates a "this file is larger than 100MB and google cannot virus check’ error.
Thus, I’m uploading the pkl file to an AWS bucket now and will try that. If you can advise how to upload direclty that would avoid all this interim ‘load here, download but get html there’ issue 
Hi LessW2020
To use use the file directly
save the pkl file in the app directory in the repository and edit the sever.py file as below.
export_file_url = ‘’
export_file_name = ‘dog_breed_34.pkl’
Cheers mrfabuous1
Unfortunately can’t upload if lareger than 25MB:
“Yowza, that’s a big file. Try again with a file smaller than 25MB.”
I’ve put it on AWS S3 so let me try to pull from that.
for reference: https://b5-100-1.s3-us-west-2.amazonaws.com/export.pkl
which does start a normal download as expected.
Thanks!
I can wget the weights with no issue from aws on my server.
However, using the same URL within server.py still yields the pickle error.
Note I tested wget to the google drive link and you just get a ‘view’ object which internally is as expected Html with first char = < : (the message boards are trying to read the html so I use * below where it’s really <
*!DOCTYPE html> … blah blah
Hi LessW2020
Is this a vision classifier?
Can you open the file and see what is inside it and compare it with the link I posted earlier.
The file I sent can be viewed in Textwrangler which is a unix version of Notepad.
It is my understanding that files with the same extension should behave the same.
Cheers mrfabulous1
ok that’s very useful info! Let me post it here in case others need it:
how to create a link from shared file
https://drive.google.com/file/d/FILE_ID/view?usp=sharing
to direct download link
https://drive.google.com/uc?export=download&id=FILE_ID
Hi everyone,
I trained a model using fastai version 1.0.52 - in the render example (https://github.com/render-examples/fastai-v3) version 1.0.51 is used. Is there a way to make this work anyway, because I don’t get my floydhub-kernel to downgrade to 1.0.51 to export the model again. Any recommendations?
Thank you so much in advance.
Hi 
piaoya hope your having a wonderful day!
As the repo uses docker, there is something you can do.
On the platform you trained your model on run “pip list” I use Google Colab so I run “!pip list”
I recommend you test your model in your notebook to make sure it still works and then save the output to a file and save it with your model.
This is because the libraries keep changing sometimes daily, sometimes without a version number change.
Once you have a your list of files, you can change the version numbers of files used in your requirements.txt in your app repository and redeploy it.
If your .pkl file is over 100mb, recently we have noticed it may not download properly from Google.
Please search this thread for posts on pip list if you need to.
Cheers mrfabulous1
Im With Render but this error
This is my environment
aiofiles==0.4.0
aiohttp==3.5.4
asyncio==3.4.3
fastai==1.0.54
https://download.pytorch.org/whl/cpu/torch-1.0.0-cp37-cp37m-linux_x86_64.whl
https://download.pytorch.org/whl/cpu/torchvision-0.3.0-cp37-cp37m-linux_x86_64.whl
numpy==1.16.3
pillow~=6.0
python-multipart==0.0.5
starlette==0.12.0
uvicorn==0.7.1
Hi Theopilus.27 hope all is well!
This is my environment
Is this the same as the environment you trained your model on.
If it isn’t you need to make sure that you have. the same version of libraries in your app as was in your training environment.
The error seems to indicate there is a problem with. the torchvision library however we need to confirm the above has been done before anything else. (See posts on pip list in this thread for further details)
Cheers mrfabulous1
Hi, I will try to deploy the generator part of CycleGan. Is there a sample code available for where input and output are both images?
Mh… using pip list I get the information on which version of fastai I am using. (I could also use import fastai; fastai.__version__ to find this out, right?) The most recent version is 1.0.60 but if I train a model on 1.0.60 I get problems with render (example runs with 1.0.52).
Is there a workaround? Downgrading fastai on colab seems not to work. Is there a upgrade from Render to expect?
