Deploying with Binder/Voila (following fastbook)

fastai fastai dev
1

Following fastbook 02_production.ipynb I wanted to clarify something:

You need the export.pkl file in your GitHub repo, correct? Otherwise, when Binder builds your app and deploys it, there won’t be any weights to load for inference.

But GitHub frowns upon large files, and will even reject them if they are too big (mine is).

So how do we get our inference model loaded?

2 Likes
2

some issue here, my export .pkl is 180Mb

3

Try gitlab or another source code repository that binder supports? My pkl file was 87mb, so worked for me

4

How did you do it?

5
6

I am having the exact same issue. Anyone had any luck finding a solution?

7

Finally, found a solution to track large files on Git .

1 Like
8

Thank you @davidsalazarvergara, the large file storage for git solved my export.pkl storage problem.

But, I have been getting a strange 404 error in upon launching my binder link, in spite of following the book’s example by the letter:

I’ve been struggling with this, have you guys gotten your binder to work?

9

I had a similar issue deploying to Binder.

I don’t know if this also applies to you, but I believe my problem was that I was training my model on a Windows machine. I tried numerous times to deploy my Windows-trained model to Binder, but it never worked.

Then I set up a VM on Google Cloud, trained my model on that instance, and then uploaded that model to my GitHub repo. That worked on my first try.

This is my first time working with Binder, so I’m obviously not familiar with it. But my impression is that it works better models built on Linux machines.

10

It seems git large files storage can be installed only in windows. I am using a VM in paperspace any idea hoy to install it there?

11

Here is my link. I have added the .pkl and .gitattributes. Still no luck!
I have also tried a number of the others on this thread and none seem to work.
  • All have 404 error
    Please let me know what is going on with binder. I am on a MacBook Pro '16.
  • Download VS Code. It is super easy to push anything to github on there
12

Hey there. I downloaded the export.pkl file from Paperspace to my WIndows laptop and then used Git LFS to upload it to the gihub repo.

1 Like
13

I put big files like models onto Google drive and load them from there.

1 Like
14

I could actually upload the .pkl via GitHub Windows desktop application

1 Like