Input:

• a Kaggle notebook with a .pkl file created (ie. the results you have from following till Lesson 2 > Turning Your Model into an Online Application > Using the Model for Inference)

Output:

• a public, sharable Hugging Face Space running your model with a nice UI

  

References

Video: https://www.youtube.com/watch?v=F4tvM4Vb3A0&t=3655s

• Blogpost with more detailed instructions: https://www.tanishq.ai/blog/gradio_hf_spaces_tutorial

Python notebook: https://github.com/fastai/fastbook/blob/master/02_production.ipynb

The video seems to do it one way (Hugging Face + Gradio), and actual text Python notebook another (Voila + Binder). Here’s my text version of the approach that the video and blogpost go over!

Some of my tweaks / disclaimers

• I’m using MacOS
• Didn’t use Bing Azure
  • Was too annoying to set up and kept running into errors on their UI. Stuck with the DuckDuckGo API used in Lesson 1.
• Used venv + pip instead of mamba

Download the .pkl you generated

(The video shows opening the notebook in “View” mode but I couldn’t figure out how to find it)

I did it this way instead: by looking in the sidebar, finding the .pkl file in the “Output” section, clicking on the 3 dots, and then hitting “More actions” to find the “Download” button

Sign up for Hugging Face